The tables below were used to generate all the graphs shown on this website.

Synthetic Sport Datasets

file aim.study sport.domain methods.model group group.description examples dataset.type dataset dataset.name data.access data.type description use.case.type use.case aim.dataset
s10439-022-02911-6 Detect head impacts using physics-informed ML models. Football Finite Element (FE) Simulation-based Use computer-generated environments, physics engines, or rule-based models to imitate athlete behavior or physical events when real data collection is limited or unsafe. Simulating training sessions, football gameplay, biomechanical impacts, movements or training loads. Synthetic GSHIKD Generated Synthetic Head-Impact Kinematic Dataset Private Tabular Synthetic head impact data generated using a physics based finite element (FE) head neck model simulating 1,065 unique impact configurations (varying velocity, orientation, ram offset, impact location). Synthetic outputs replicate instrumented mouthguard kinematic signals. 6 DOF linear and angular accelerations, velocities, and displacements. Injury Head Impact classification modelling in American football using sensors.
  1. Generate realistic impact kinematics to address imbalance between true impacts and false positives in mouthguard data. 2. Enable training of detectors to identify dangerous head impacts without full reliance on video review.
1-s2.0-S026288562500277X-main Review and classify advances in basketball action recognition (BAR) using AI. Swimming GANimator GAN/NN-based Generate realistic synthetic data using Neural Network or optimised version of it as GAN with two competing neural networks one creates data, the other judges realism until the synthetic samples resemble the real ones. Creating synthetic motion capture, wearable-sensor, or video data for training, recovery or performance analysis. Synthetic SwimXYZ Large-scale dataset of synthetic swimming motions and videos Public Access Video Large-scale synthetic swimming motion dataset (2D/3D). Movement Human motion analysis and pose estimation underwater.
  1. Provide synthetic 2D/3D videos with joints and positions of each joint. 2. Enable training for underwater human pose estimation.
1-s2.0-S026288562500277X-main Review and classify advances in basketball action recognition (BAR) using AI. Basketball PoseNet/IdentityNet/SkinningNet GAN/NN-based Generate realistic synthetic data using Neural Network or optimised version of it as GAN with two competing neural networks one creates data, the other judges realism until the synthetic samples resemble the real ones. Creating synthetic motion capture, wearable-sensor, or video data for training, recovery or performance analysis. Synthetic NBA2K Reconstructing NBA Players Public Access Image Synthetic basketball dataset derived from NBA2K19 (3D images, meshes, textures, poses, camera parameters). Movement 3D reconstruction, pose estimation, and player tracking in basketball.
  1. Introduce the SoccER synthetic data generator for soccer event analysis.
1-s2.0-S026288562500277X-main Review and classify advances in basketball action recognition (BAR) using AI. Soccer Unreal Engine Simulation-based Use computer-generated environments, physics engines, or rule-based models to imitate athlete behavior or physical events when real data collection is limited or unsafe. Simulating training sessions, football gameplay, biomechanical impacts, movements or training loads. Synthetic SoccerSynth Synthetic Dataset for Soccer Player Detection Public Access Image Synthetic soccer player dataset for detection and tracking. Movement Player/ball detection under varied lighting, blur, and in-game scenarios.
  1. Provide controllable synthetic scenes for training robust detection models. 2. Simulate motion blur to model real world conditions.
Monocular_3D_Human_Pose_Estimation_for_Sports_Broadcasts_using_Partial_Sports_Field_Registration Extract 3D poses from 2D broadcast video using field calibration. Running Unreal Engine Simulation-based Use computer-generated environments, physics engines, or rule-based models to imitate athlete behavior or physical events when real data collection is limited or unsafe. Simulating training sessions, football gameplay, biomechanical impacts, movements or training loads. Synthetic SMDRD Synthetic Middle-Distance Running Dataset Public Access Video A fully synthetic dataset of 10,571 rendered images generated in Unreal Engine 5 using Mixamo and MetaHumans. Includes ground-truth 3D joint locations, 2D joints, and complete camera calibration for middle-distance running on a 400m track. Contains 31 sequences with variations in athlete body type, running technique, and camera motion (pan/tilt/zoom), replicating real broadcast conditions. Movement Evaluation of monocular 3D pose estimation methods in sports broadcasts.
  1. Provide a benchmark synthetic dataset to test 3D pose estimation accuracy in running. 2. Replicate broadcast style camera setups to study zoom, angle, and distance effects. 3. Enable development of 3D kinematic extraction pipelines for coaching, performance analysis, and biomechanical assessment. 4. Provide high quality ground truth where real MoCap is unavailable.
978-3-031-72353-7_12 Compare synthetic time-series generators for athlete fatigue data. Running/Triathlon/Ski mountaineering K-medoids/Transition Probabilities/VAE/TimeGAN/TimeGrad. GAN/NN-based Generate realistic synthetic data using Neural Network or optimised version of it as GAN with two competing neural networks one creates data, the other judges realism until the synthetic samples resemble the real ones. Creating synthetic motion capture, wearable-sensor, or video data for training, recovery or performance analysis. Synthetic SAMD Synthetic Athlete Monitoring Dataset Private Tabular Synthetic multivariate time series generated from 8-day sequences of real athlete metrics (sleep quality, mood, Foster load, and O2score). Built from 5 endurance athletes practicing ski mountaineering, triathlon, and trail running. Performance Athlete monitoring augmentation for fatigue prediction, training response modelling, and recovery analysis.
  1. Create realistic synthetic athlete monitoring data to compensate for extremely small datasets. 2. Preserve temporal dynamics needed for predicting O2score trends. 3. Provide synthetic scenarios to study athlete specific variability in mood, sleep, training load, and physiological response without invasive measurements.
fspor-7-1607600 Predict performance attenuation using synthetic athlete data. Football Tabular Variational Autoencoders (TVAE) GAN/NN-based Generate realistic synthetic data using Neural Network or optimised version of it as GAN with two competing neural networks one creates data, the other judges realism until the synthetic samples resemble the real ones. Creating synthetic motion capture, wearable-sensor, or video data for training, recovery or performance analysis. Synthetic GFPD Generated Gaelic Football Performance Dataset Public Access Tabular Synthetic tabular dataset generated using a Tabular Variational Autoencoder (TVAE) trained on neuromuscular, perceptual, biochemical, GPS, and anthropometric data from 41 Gaelic football players. Performance Performance attenuation prediction in Gaelic football using athlete monitoring augmentation.
  1. Expand Gaelic football monitoring datasets to improve prediction of fatigue related performance decline. 2. Generate realistic neuromuscular, biochemical, and perceptual patterns replicating match play demands. 3. Support coaching decisions on load management, recovery, and readiness.
ico-2021-synthetic-data-augmentation-of-cycling-sport-training-datasets Generate synthetic cycling sessions to enrich training archives. Cycling Activity Recognition Model (ARM)/Activity Sequence Tracking (AST), SportyDataGen GAN/NN-based Generate realistic synthetic data using Neural Network or optimised version of it as GAN with two competing neural networks one creates data, the other judges realism until the synthetic samples resemble the real ones. Creating synthetic motion capture, wearable-sensor, or video data for training, recovery or performance analysis. Synthetic SCTS Synthetic Cycling Training Sessions Private Tabular Synthetic cycling training sessions generated by increasing session intensity (HR) on the most demanding rides from a real archive. Uses a Training Stress Measure (TSM) and regression-based prediction of speed to create new, more intense sessions beyond those present in the athlete real training history. Performance Cycling training modelling for session planning, load progression, and intensity performance relationships.
  1. Produce more intense synthetic cycling sessions (higher HR, speed, TSM) to satisfy progressive overload. 2. Expand training archives so AI systems can recommend higher load sessions. 3. Capture how increases in HR intensity influence training stress. 4. Provide realistic uncommon sessions to train models on unseen scenarios.
ijspp-article-p1213 Demonstrate synthetic data for secure sharing in sports science. Rugby Synthpop (CART) Statistical-based Produce new samples by reproducing statistical patterns or learned variable relationships using mathematical models rather than neural networks. Generating realistic athlete performance tables, expanding small datasets, or privacy preserving data sharing through fitting mathematical relationships. Synthetic GTSRD Generated Team-Sport Load and Recovery Datasets Public Access Tabular Synthetic datasets created using the synthpop R package from two real datasets. (1) weekly preseason training load data from academy rugby league players (n = 28), including GNSS, gym load, and RPE; (2) fatigue recovery markers from male and female team-sport athletes (n = 22), including soreness, CMJ, and sprint performance across Pre, Post1, Post2. Performance Privacy preserving sport science analysis for load monitoring, fatigue profiling, and performance modelling.
  1. Enable exploratory sport science analysis when data cannot be shared. 2. Provide synthetic team sport datasets for teaching, hypothesis generation, and collaborations. 3. Preserve key performance relationships (load fatigue, RPE trends, CMJ changes).
s40279-025-02221-6 Demonstrate synthetic data creation using R synthpop for sport science. Football Synthpop (CART) Statistical-based Produce new samples by reproducing statistical patterns or learned variable relationships using mathematical models rather than neural networks. Generating realistic athlete performance tables, expanding small datasets, or privacy preserving data sharing through fitting mathematical relationships. Synthetic SPDAM Sport-Performance Data for Athlete Monitoring Public Access Tabular Synthetic multivariate performance dataset generated using machine learning based tabular models trained on real athlete-monitoring variables (training load). The synthetic data replicate realistic distributions and correlations of high performance sport measures. Performance Sport performance prediction, readiness classification, and training load decision support.
  1. Enable testing and comparison of models predicting readiness, fatigue, or performance without proprietary data. 2. Preserve key load response relationships for simulation of decision support scenarios. 3. Provide a safe dataset for researchers and practitioners to validate modelling tools before deployment.
frai-05-988113 Improve player jersey number detection via synthetic datasets. Football CNN GAN/NN-based Generate realistic synthetic data using Neural Network or optimised version of it as GAN with two competing neural networks one creates data, the other judges realism until the synthetic samples resemble the real ones. Creating synthetic motion capture, wearable-sensor, or video data for training, recovery or performance analysis. Synthetic SC2DSJND Simple2D and Complex2D Synthetic Jersey Number Datasets Available on request Image Two synthetic datasets created to address extreme class imbalance and low-sample availability in jersey number detection for American football. Simple2D generates two digit numbers using Seahawk like fonts, colors, and controlled augmentations. Complex2D overlays those numbers on COCO images to simulate realistic noisy backgrounds. Player Jersey number recognition for American football to improve player identification.
  1. Generate realistic synthetic jersey numbers mimicking team fonts and color palettes. 2. Provide controlled variations (lighting, noise, backgrounds) to help models generalize to sideline and end-zone practice videos.
Machine_learning_based_Synthetic_Data_Generation_using_Iterative_Regression_Analysis Create pseudo-real data to boost model accuracy. Soccer Iterative Regression Statistical-based Produce new samples by reproducing statistical patterns or learned variable relationships using mathematical models rather than neural networks. Generating realistic athlete performance tables, expanding small datasets, or privacy preserving data sharing through fitting mathematical relationships. Synthetic GFPAD Generated FIFA Player Attribute Dataset Private Tabular Fully synthetic soccer-performance dataset generated using iterative regression models trained on FIFA player attributes (Ball Control, Dribbling, Special, Short Passing, Long Passing). Synthetic values are created by feeding random vectors through sequential regression models with added noise and randomness validation (Runs Test, Chi-Square). Player Player attribute expansion for soccer analytics, including technical-ability prediction and player classification.
  1. Provide additional samples to improve model performance for predicting football skill ratings. 2. Enable scouting, ranking, and performance-projection modelling. 3. Create synthetic scenarios showing how ball control, passing, and dribbling evolve statistically.
1-s2.0-S2352711020303253-main Introduce the SoccER synthetic data generator for soccer event analysis. Soccer Computer graphics simulation engine Simulation-based Use computer-generated environments, physics engines, or rule-based models to imitate athlete behavior or physical events when real data collection is limited or unsafe. Simulating training sessions, football gameplay, biomechanical impacts, movements or training loads. Synthetic SoccER Synthetic positional data extracted from the Game Football engine Public Access Tabular Fully synthetic soccer dataset generated from 8 complete simulated matches using a modified Gameplay Football engine. Includes spatio temporal player/ball coordinates, event labels (tabular), video frames, bounding boxes, and CVAT annotations. Tactical Soccer event recognition for passes, tackles, shots, fouls, possession, deflections, and goal actions.
  1. Provide a realistic, fully annotated soccer environment to study event patterns without relying on expensive real game data. 2. Enable controlled evaluation of event recognition algorithms across all major soccer actions and interactions. 3. Generate tactical behaviours, ball movement, and player interactions using dense spatio temporal ground truth.
Data_Augmentation_and_Neural_Network_for_American_Football_Formation_Recognition Automate offensive play recognition with augmented data. Football Multilayer perceptron (MLP) GAN/NN-based Generate realistic synthetic data using Neural Network or optimised version of it as GAN with two competing neural networks one creates data, the other judges realism until the synthetic samples resemble the real ones. Creating synthetic motion capture, wearable-sensor, or video data for training, recovery or performance analysis. Synthetic SAFFD Synthetic American Football Formation Dataset Private Tabular Synthetic dataset created through structured data augmentation applied to 25 common offensive play formations. Each formation consists of 11 offensive players X and Y coordinates, expanded using position specific displacement ranges, rotation, and scaling to produce 5,000 realistic formation samples. Tactical Recognition of offensive football formations from player location data.
  1. Capture variation in receiver, tight end, backfield, and line positions to simulate real in-game pre snap diversity. 2. Support automated formation recognition systems for strategy analysis. 3. Enable robust neural network training by adding controlled noise (rotations, shifts, scaling).
main Build synthetic labeled datasets for sports intelligence. Handball/Basketball ML GAN/NN-based Generate realistic synthetic data using Neural Network or optimised version of it as GAN with two competing neural networks one creates data, the other judges realism until the synthetic samples resemble the real ones. Creating synthetic motion capture, wearable-sensor, or video data for training, recovery or performance analysis. Synthetic SHBGSD Synthetic Handball and Basketball Game-Situation Dataset Public Access Video Synthetic, normalized JSON dataset derived from real video frames of 2105 handball clips and 383 basketball clips. Contains 308,805 handball frames and 56,578 basketball frames labeled into seven game-situation classes (left/right attack, counterattack, penalty, timeout). Each synthetic frame encodes player, referee, and ball positions, player velocities, and court-region markers projected into a standardized unified space. Tactical Game situation classification in handball and basketball using spatio temporal representations.
  1. Provide a large labelled synthetic dataset of indoor team sport game dynamics. 2. Support tactical pattern learning across attacks, counters, and penalties. 3. Supply high resolution positional data across matches and arenas. 4. Enable research on game situation recognition without video. 5. Support applications such as sports analytics, strategy optimisation, and match retransmissions.

Real Sport Datasets

column study.title dataset.name dataset dataset.type methods.model use.case.type use.case aim.dataset valid.data total.score synthetic.generation country year.start year.end year.range population.age.range population.type population.sex population sample.overall sample.raw sample.size study.design sport.type sports.covered data.type variables.collected literature.category
NCAA-ISP Methods of the National Collegiate Athletic Association Injury Surveillance Program, 2014-2015 Through 2018-2019 National Collegiate Athletic Association Injury Surveillance Program NCAA-ISP Real Datalys Center. Injury Data collection initiative designed to track and analyze medical illnesses and injuries that result from sport participation.
  1. Track and analyse injuries and medical illnesses related to sport participation. 2. Maintain a national collegiate sports injury database to inform prevention policies. 3. Support collaboration between athletic trainers, the NCAA, and the Datalys Center to improve injury reporting.
Yes 4.15 No US 2005 2019 2005-2019 High school Athlete Both U.S. high school athletes across sanctioned sports 500000 500000 Aprox. 10.5 million athlete exposures Retrospective Multiple Football (boys), Soccer (boys/girls), Volleyball (girls/boys), Basketball (boys/girls), Wrestling, Baseball, Softball, Field Hockey, Gymnastics, Ice Hockey, Lacrosse (boys/girls), Swimming/Diving (boys/girls), Track & Field (boys/girls), Cheerleading, Cross Country (boys/girls), Tennis (boys/girls) Medical Record
  1. Athlete demographics (sex, age, class year, height, weight) 2. Injury info (body part, type, mechanism, severity, outcome, surgery, RTP time) 3. Event info (practice vs competition, level of play, season, surface, weather) 4. Equipment (helmet, pads, braces, mouthguard, protective eyewear) 5. Sport-specific variables (position, play activity, injury mechanism) 6. Environmental and contextual data (weather, temperature, humidity)
Statistical
MTS-5 A novel multivariate time series dataset of outdoor sport activities Sport Activity Multivariate Time Series Dataset 5 MTS-5 Real Wearable devices. Performance Outdoor sport activity recognition and performance analysis.
  1. Provide multivariate time-series of heart rate, speed, and altitude recorded in real outdoor environments. 2. Support sports performance analysis across walking, running, skiing, roller-skiing, and biking. 3. Facilitate research on physiological–environmental interactions such as altitude, speed, and heart rate dynamics.
Yes 4.30 Yes. Augmented by concatenating five consecutive one-minute segments from the same activity Russia 2023 2024 2023-2024 Adult Athlete Male One non-competitive adult male athlete performing various outdoor activities 228 1140 228 full-length outdoor activities; 1140 one-minute segments after preprocessing Longitudinal Multiple Walking, Running, Skiing, Roller-Skiing, Biking Physiological
  1. Heart Rate (bpm) 2. Speed (m/s, GPS-derived) 3. Altitude (m, barometric pressure–based). 4. Metadata: sport category, device type (Garmin Forerunner 920XT, Garmin Vivosport). 5. Derived metrics (segment averages, standard deviations, standardized z-scores).
Statistical
Pre-SCoV2 Comprehensive Dataset on Pre-SARS-CoV-2 Infection Sports-Related Physical Activity Levels, Disease Severity, and Treatment Outcomes: Insights and Implications for COVID-19 Management Pre-SARS-CoV-2 Infection Sports-Related Physical Activity Levels, Disease Severity, and Treatment Outcomes Dataset Pre-SCoV2 Real Online self-report questionnaire. Injury To examine how pre-infection physical activity levels relate to COVID-19 disease severity and treatment outcomes.
  1. Provide harmonized physical activity and COVID-19 severity categories for analysis. 2. Enable evaluation of associations between PA levels and treatment patterns.
Yes 2.70 No Greece NA 2023 2023 Adult Athlete Both Adults (aged 18–70+) with confirmed SARS-CoV-2 infection within 30–40 days prior to participation 5829 5829 5,829 participants (1,962 males; 3,867 females) Cross-sectional Multiple Active-Q questionnaire: aerobics/cardio, weightlifting, jogging/running, athletics, spinning/cycling, swimming, team ball sports (soccer, basketball, volleyball, floorball), dance, horseback riding, ice hockey, skiing, martial arts, boxing/wrestling, yoga/Pilates/Tai chi, tennis/badminton/table tennis, squash, sailing/rowing, motorsports, rock climbing Survey
  1. Demographics (age, sex, education, region, ethnicity). 2. Anthropometrics (height, weight, BMI). 3. Sports-related physical activity (MET-min/week, frequency, duration). 4. Medical conditions (19 high-risk categories). 5. Vaccination status (type, doses, intervals). 6. Reinfection frequency. 7. COVID-19 disease severity (5-level scale). 8. Treatment category (5-level scale).
Statistical
ICAD Cross-sectional and longitudinal associations of active travel, organised sport and physical education with accelerometer-assessed moderate-to-vigorous physical activity in young people: the International Children’s Accelerometry Database International Children’s Accelerometry Database ICAD Real Self-reported physical activity domains. Performance To understand daily moderate-to-vigorous physical activity (MVPA) in children and adolescents.
  1. Quantify cross-sectional links between domain-specific activities and MVPA. 2. Assess whether baseline activity domains predict future changes in MVPA. 3. Inform policy and interventions on which activity domains contribute most to youth physical activity.
Yes 2.55 No Europe, US, Brazil, Australia 2008 2017 2008-2017 Adolescent Multiple Both Children and adolescents (5–18 years) from international cohort studies 47000 47000 >47,000 participants from 20+ contributing studies (variable by data type) ALL Fitness Free-living physical activity and sedentary behavior across daily life (not specific sports, but includes intensity classifications from sedentary to vigorous) Accelerometer
  1. Accelerometer variables: counts per minute (CPM), wear time (daily/hourly), sedentary/light/moderate/vigorous activity durations, bouted and accumulated intensities, Evenson/Pate/Block cut-point derived variables. 2. Non-accelerometer variables: demographics, anthropometrics (height, weight, waist, skinfolds), blood pressure, glucose, insulin, lipids, diet, home/family, and school-related data.
Statistical
LLBD A Quantitative Assessment Grading Study of Balance Performance Based on Lower Limb Dataset Lower Limb Dataset for Quantitative Balance Assessment LLBD Real Inertial measurement units (IMUs) collecting lower-limb 3D motion signals. Movement To measure, classify, and quantify human balance during controlled motion tasks using wearable sensor data.
  1. Support objective grading of balance ability using sensor-based motion features. 2. Provide data for designing balance-training tools, screening protocols, and sport-performance assessments.
Yes 3.65 No China NA 2022 2022 Adult Multiple Both College students (mean age 25.35 ± 2.35 years) screened for normal motor function and balance 800 20 800 trials, 20 subjects × 4 movement types × 10 repetitions (7-sensor IMU data at 100 Hz, multiple frames per trial) Cross-sectional (Repeated measures) Fitness Static standing (upright), walking (6-step gait), squatting, and BOSU-ball squatting representing balance and coordination tasks Accelerometer
  1. 3D acceleration (x,y,z) 2. 3D angular velocity (x,y,z). 3. Quaternion (w,x,y,z) 4. Velocity (x,y,z). 5. Position (x,y,z). 6. BVH skeleton data (joint rotation angles and bone lengths). 7. Derived features (RMS, kurtosis, skewness, ApEn, total energy, mean motion period, peak power).
Statistical
UCF-SAD Exploring the potential of deep learning techniques for analyzing athlete movements in competitive athletics sports UCF Sports Action Dataset UCF-SAD Real Collected from broadcast television channels. Injury To predict professional football players’ injury risk and identify key factors contributing to future injury probability.
  1. Provide integrated biomarker, training-load, and medical-history features for injury-risk modeling. 2. Enable development and comparison of machine-learning predictive algorithms.
Yes 3.10 No US, UK NA 2008 2008 Adult Athlete Both Professional athletes and sport participants captured from real TV broadcasts 150 10 150 video clips, 10 sports actions (720×480 resolution) Cross-sectional Multiple Diving, Golf Swing, Kicking, Lifting, Riding Horse, Running, Skateboarding, Swinging Sideways, Swinging on Bench, Walking Video
  1. RGB frames (720×480). 2. Human bounding boxes. 3. Action labels (10 classes). 4. Temporal segmentation. 5. Gaze annotation.
GAN-based
OSD Exploring the potential of deep learning techniques for analyzing athlete movements in competitive athletics sports Olympic Sports Dataset OSD Real Collected from Youtube. Injury To predict professional football players’ injury risk and identify key factors contributing to future injury probability.
  1. Provide integrated biomarker, training-load, and medical-history features for injury-risk modeling. 2. Enable development and comparison of machine-learning predictive algorithms.
Yes 3.15 No International NA 2010 2010 Adult Elite Both Olympic athletes performing official Olympic events 800 16 800 video clips, 16 Olympic sports; 50 videos per sport class Cross-sectional Multiple High Jump, Long Jump, Triple Jump, Pole Vault, Discus Throw, Hammer Throw, Javelin Throw, Shot Put, Basketball Layup, Bowling, Tennis Serve, Platform Diving, Springboard Diving, Snatch, Clean & Jerk, Gymnastics Vault Video
  1. Spatiotemporal motion features (3D Harris interest points). 2. Histogram of Gradients (HoG). 3. Histogram of Optical Flow (HoF). 4. Temporal segments and anchor points for each motion classifier.
GAN-based
HTHARD Hang-Time HAR: A Benchmark Dataset for Basketball Activity Recognition Using Wrist-Worn Inertial Sensors Hang-Time HAR (Human Activity Recognition) Dataset for Basketball Activity HTHARD Real Wrist-worn sensors collecting multi-device inertial data. Movement Collect synchronised multi-sensor inertial data during basketball activities (locomotion, shots, layups) and annotate them with multi-tier ground-truth labels.
  1. Provide high-resolution IMU signals for basketball-specific HAR research. 2. Enable benchmarking of deep learning models for classifying complex basketball movements.
Yes 3.15 No Germany, US NA 2021 2021 Adult Athlete Both Professional basketball players (21 male, 3 female); ages 18–39; mixed skill levels (16 expert, 8 novice); 13 players from Germany and 11 from USA 5030378 24 24 players × multi-session recordings (aprox. 27.7 hours of labeled data), 5,030,378 samples Cross-sectional Basketball Basketball (including both FIBA and NBA rules) Accelerometer
  1. 3D acceleration (x, y, z). 2. Timestamp (ms). 3. Participant metadata (ID, age, sex, hand dominance, height, weight, experience level). 4. Four annotation layers: Coarse (session type), Basketball (activity), Locomotion (movement type), In/Out (court participation).
Statistical
NATOPS Applying Deep Learning-Based Human Motion Recognition System in Sports Competition NATOPS Video Dataset NATOPS Real Video collection from small hand movements in 24 sports fields. Movement To understand human motion recognition system for sports competitions, using video data to detect, track, and classify athletic movements.
  1. Recognize micromotions and fine-grained hand/ball interactions in sports. 2. Fuse manual (trajectory, static features) and deep-learning RGB/time features for more robust tactical analysis.
Yes 1.45 No China NA 2022 2022 Adult Athlete NA Adults performing standardized hand gestures and micromotions 480 20 480 total actions, 20 gesture categories × 24 hand movements per category Cross-sectional Multiple Micromotions and gesture actions in basketball and volleyball Video
  1. RGB color frames. 2. Temporal motion templates (64×48 frames). 3. Fuzzy membership functions. 4. Manual (IDT) and deep learning (CNN) feature descriptors.
GAN-based
RBD Multi-Output Sequential Deep Learning Model for Athlete Force Prediction on a Treadmill Using 3D Markers Running Biomechanics Dataset (RBDS) RBD Real Inertial sensors, motion tracking, EMG signals for collection. Movement To capture synchronised multimodal biomechanical signals for the estimation and evaluation of elbow and shoulder joint torques during different dynamic actions.
  1. Provide multimodal sensor data for modelling human upper-limb kinetics. 2. Support rehabilitation, ergonomics, and sports-performance assessment by quantifying muscular and joint loading.
Yes 2.90 No Brazil NA 2017 2017 Adult Athlete Male Runners (≥20 km/week, 5 min/km pace, familiar with treadmill running) 168 20 28 participants × 3 speeds × both limbs, 168 raw motion and force datasets (aprox. 12 files per subject) Cross-sectional (Repeated measures) Running Running biomechanics (treadmill running) Biomechanical
  1. 3D marker coordinates (96 markers). 2. Ground reaction forces (Fx, Fy, Fz). 3. Center of pressure (COPx, COPy, COPz). 4. Joint torques and powers (hip, knee, ankle). 5. Demographics, foot-strike patterns, running habits, muscle strength, and flexibility metadata.
Statistical
CAIFSD Prevalence and impact of chronic ankle instability in female sport: a cross-sectional study Chronic Ankle Instability in Female Sport Dataset (CAI-FSD) CAIFSD Real Online cross-sectional survey. Injury Assessment of ankle health, instability, and functional impact in female participants of netball, soccer, basketball, and volleyball.
  1. Identify the prevalence of chronic ankle instability in female athletes. 2. Evaluate its impact on ankle function and quality of life. 3. Provide sport and sex specific evidence to guide rehabilitation and injury prevention.
Yes 2.70 No Australia, New Zealand, UK, US NA 2024 2024 Adult Athlete Female Female athletes (≥ 18 years) participating in netball, soccer, basketball, volleyball for at least 12 months 258 578 578 survey responses received; 258 complete datasets used for analysis (Australia 27%, n = 106; New Zealand 19%, n = 75; UK 44%, n = 170; USA 7%, n = 29) Cross-sectional Multiple Netball, soccer, basketball, and volleyball Survey
  1. Demographics (age, height, mass, country, level of play, years in sport). 2. Ankle injury history (type, side, frequency, fracture vs sprain). 3. Cumberland Ankle Instability Tool (CAIT) scores. 4. Foot and Ankle Ability Measure Sport (FAAM-S) scores. 5. Health-Related Quality of Life (HRQOL-14) scores. 6. Classification (CAI, Potential Coper, No Injury).
Statistical
Sports-1M Big Data and Deep Learning-Based Video Classification Model for Sports Sports 1Million Dataset Sports-1M Real Collected from YouTube Topics API. Tactical Automatic classification of sports videos.
  1. Identify and classify complex sports movements from video data. 2. Support automatic description and semantic labeling of sports actions.
Yes 2.95 No International NA 2014 2014 Adult Multiple Both Public sports videos on YouTube representing a broad range of athletic activities 1000000 1000000 1 million YouTube videos belonging to 487 sports classes (aprox.1 000–3 000 videos per class) Cross-sectional Multiple 487 sport categories organized in a hierarchical taxonomy: Aquatic Sports, Team Sports, Winter Sports, Ball Sports, Combat Sports, Sports with Animals, etc.; includes 6 types of bowling, 7 of American football, 23 of billiards Video
  1. Video class label (487 sports categories). 2. YouTube metadata (title, description, tags). 3. RGB video frames. 4. Sampled 10-frame clips. 5. Train/validation/test splits. 6. Automatically generated annotations (weakly labeled).
Statistical
SKMD Applying Deep Learning and Computer Vision Techniques for an e-Sport and Smart Coaching System Using a Multiview Dataset: Case of Shotokan Karate Multiview Shotokan Karate Dataset (SKMD) SKMD Real Multiview videos of a 6th-Dan Shotokan Karate athlete. Movement To enable automated detection, recognition, classification, and scoring of Karate techniques using deep learning for smart coaching and e-sport systems.
  1. Enable consistent classification of Karate techniques across multiple camera views. 2. Provide reliable reference movement patterns for automated scoring and feedback.
Yes 2.35 No Morocco NA 2021 2021 Adult Athlete Male 6th-Dan Shotokan Karate expert (coach) performing 8 basic movements recorded for pose estimation; dataset designed for training intelligent e-coaching systems 24 1 24+ video segments aprox. used for model training and testing; 8 Shotokan Karate movements × 3 camera views (Front, Right, Left) × multiple repetitions per view Cross-sectional Karate Shotokan Karate Video
  1. 2D body keypoints (OpenPose/FastPose: up to 135 joints). 2. 3D body coordinates (via VIBE reconstruction). 3. Pose estimation confidence maps. 4. Movement type labels (8 classes: Gedan Barei, Tsuki Chudan, Tsuki Jodan, Soto Uke, Shuto Uke, Mae Giri, Mawashi Giri, Yoko Giri). 5. View identifier (Left, Right, Front). 6. Model parameters and training logs (LSTM, ST-GCN).
GAN-based
FFTSC-10Y Nonoperative Treatment of Finger Flexor Tenosynovitis in Sport Climbers-A Retrospective Descriptive Study Based on a Clinical 10-Year Database Finger Flexor Tenosynovitis in Sport Climbers 10-Year Clinical Database FFTSC-10Y Real Retrospective 10 year clinical database with ultrasound-based diagnosis and follow-up questionnaires. Injury Evaluate injury characteristics and outcomes of conservative treatment for finger flexor tenosynovitis in sport climbers.
  1. Describe injury patterns, triggers, and therapy contents in climbers. 2. Assess the effectiveness of nonoperative treatment in reducing pain and restoring climbing function. 3. Identify baseline factors associated with symptom duration and recovery.
Yes 2.15 No Switzerland 2010 2019 2010-2019 Adult Athlete Both Adult sport climbers (≥ 18 years) diagnosed with finger flexor tenosynovitis related to climbing 65 65 65 sport climbers (49 male, 16 female; mean age 34.1 years) Retrospective Climbing Rock climbing / sport climbing Survey
  1. Demographics (age, sex, climbing level, years of experience). 2. Ultrasound pulley thickness (A2, A4). 3. Pain intensity (VAS). 4. Therapy components (modelling clay, compression fingerling, taping, ergotherapy, medication, load reduction). 5. Injury triggers (hard training, crimping, repeated moves). 6. Functional outcomes (SANE score, return to climbing level).
Statistical
NEISS Musculoskeletal injuries during trail sports: Sex- and age-specific analyses over 20 years from a national injury database National Electronic Injury Surveillance System (NEISS) NEISS Real Retrospective extraction of 20 years of trail-sport injury records. Injury Identify musculoskeletal injury patterns in hikers, trail runners, and mountain bikers across age and sex groups.
  1. Describe how injury rates differ by sex, age, and trail-sport type. 2. Identify which diagnoses and body regions are most affected in trail sports. 3. Support injury-prevention planning by showing long-term trends in trail-sport injuries.
Yes 3.00 No US 2002 2021 2002-2021 Adult Multiple Both U.S. emergency department patients with musculoskeletal injuries sustained during trail sports (hiking, trail running, or mountain biking) 9835 9835 9,835 injuries cases recorded Retrospective Multiple Hiking, Trail Running, Mountain Biking Medical Record
  1. Age, sex, race, ethnicity. 2. Diagnosis code (amputation, concussion, contusion, crushing, dislocation, fracture, laceration, nerve damage, puncture, strain/sprain). 3. Body part injured. 4. Product code. 5. Injury narrative. 6. Statistical weighting variable for national estimates.
Statistical
NHL-ATR Outcomes Following Achilles Tendon Ruptures in the National Hockey League: A Retrospective Sports Database Study National Hockey League Achilles Tendon Injury Dataset NHL-ATR Real Retrospective extraction of Achilles tendon rupture cases from four publicly available NHL injury databases in websites. Injury Evaluate incidence, mechanisms, treatment strategies, return to play timelines, and performance outcomes after tendon rupture in NHL athletes.
  1. Quantify how often Achilles tendon ruptures occur in NHL players. 2. Describe mechanisms of rupture and patterns of management. 3. Assess how these injuries influence return to play rates and post-injury performance.
Yes 2.10 No Canada, US 2008 2024 2008-2024 Adult Athlete Male Professional male ice hockey players participating in the National Hockey League (NHL) 15 15 15 confirmed Achilles tendon rupture cases across 16 seasons Retrospective Ice Hokey Ice hockey (National Hockey League) Medical Record
  1. Player demographics (age, position, BMI, years active). 2. Career and season-level performance metrics (games played, goals, assists, points, plus/minus, penalty minutes, power play/short-handed stats, shots, time on ice, shooting percentage, faceoff percentage). 3. Injury characteristics (mechanism: non-contact/laceration; timing: in-season/off-season). 4. Treatment method (surgical vs non-surgical). 5. Return to play metrics (RTP rate, time missed, recovery months).
Statistical
MTFnFD AI-Assisted Fatigue and Stamina Control for Performance Sports on IMU-Generated Multivariate Times Series Datasets Multivariate Time Series Data of Fatigued and Non-Fatigued Running from Inertial Measurement Units MTFnFD Real IMU-based multivariate time-series collection. Performance Dataset created to monitor biomechanical changes, quantify fatigue and stamina prediction in sports performance.
  1. Identify biomechanical markers associated with fatigue and stamina. 2. Enable real-time monitoring and personalised training adjustments based on sensor-derived performance metrics.
Yes 2.15 No Ireland NA 2023 2023 Adult Athlete NA Adult runners, injury-free 19 19 19 participants; multiple runs (400-m pre-fatigue and post-fatigue) Cross-sectional Running Running Accelerometer
  1. Participant ID. 2. Fatigue label (F/NF). 3. Time-series of acceleration (ax, ay, az). 4. Angular velocity (wx, wy, wz). 5. Magnetic orientation (mx, my, mz). 6. Derived composite magnitudes (acceleration, gyroscope). 7. Stamina/fatigue prediction labels.
GAN-based
TeamTrack TeamTrack: A Dataset for Multi-Sport Multi-Object Tracking in Full-pitch Videos TeamTrac Dataset for Multi-Sport Multi-Object Tracking in Full-Pitch Videos TeamTrack Real Full multi-view video capture (drone top-view and fisheye side-view) with manual and interpolated annotations. Tactical To provide a large-scale benchmark for evaluating multi-object tracking in complex team-sport scenarios.
  1. Enable development of tracking models that handle players with similar appearance. 2. Support evaluation of MOT performance under occlusion, dense formations, and fast motion.
Yes 4.50 No Japan, China NA 2023 2023 Adult Athlete Both University-level male and female athletes competing in soccer, basketball, and handball games 279900 279900 > 279,900 annotated frames and 4,374,900 bounding boxes from 150 minutes of match recordings Cross-sectional Multiple Soccer, Basketball, Handball Video
  1. Bounding box coordinates (x, y, width, height). 2. Frame number. 3. Object ID (persistent player tracking). 4. Sport type and camera perspective (side or top view). 5. Team labels (in SportsLabKit format). 6. Motion metrics (IoU, trajectory positions).
GAN-based
FineSports FineSports: A Multi-person Hierarchical Sports Video Dataset for Fine-grained Action Understanding Multi-Person Hierarchical Sports Video Dataset for Fine-Grained Action Understanding FineSports Real Game videos collected from the NBA official replay archive. Tactical To build a benchmark that enables accurate analysis of fine-grained movements and multi-player interactions in basketball videos.
  1. It supports evaluating fine-grained action recognition models. 2. It enables development of spatial-temporal action localization systems.
Yes 3.30 No China NA 2024 2024 Adult Athlete Male Professional basketball players from the NBA 10000 10000 10,000 annotated NBA video samples; 16,000 action instances; 123,000 spatial bounding boxes; 32,096 temporal boundaries Cross-sectional Basketball Basketball Video
  1. Bounding box coordinates per player per frame. 2. Player ID and role (ball handler, teammate, opponent). 3. Action type (12 categories). 4. Sub-action type (52 fine-grained procedures). 5. Frame number. 6. Temporal boundaries of actions.
GAN-based
DeepSportradar-v1 DeepSportradar-v1: Computer Vision Dataset for Sports Understanding with High Quality Annotations Computer Vision Dataset for Sports Understanding with High Quality Annotations DeepSportradar-v1 Real High-resolution basketball game footage captured with camera. Tactical Provide real-world multi-task computer-vision data for benchmarking and developing basketball analytics models.
  1. It enables precise estimation of ball position, camera parameters, and player boundaries in real games. 2. It supports player tracking and identity recognition across diverse arenas.
Yes 2.80 No France NA 2022 2022 Adult Athlete Male Professional male basketball players, coaches, and referees from the French LNB Pro A league 423 37 37 games across 15 arenas; 223 training, 37 validation, 64 testing images for segmentation; 99 video sequences and 18,703 cropped player thumbnails for re-identification Cross-sectional Basketball Basketball Image
  1. Ball 3D coordinates. 2. Camera calibration matrices (intrinsic/extrinsic). 3. Human segmentation masks. 4. Player thumbnails with identity labels. 5. Arena labels and camera metadata.
GAN-based
MultiSports MultiSports: A Multi-Person Video Dataset of Spatio-Temporally Localized Sports Actions Multi-Person Video Dataset of Spatio-Temporally Localized Sports Actions MultiSports Real Manually annotated soccer broadcast video frames Tactical Support accurate and real-time detection of the soccer ball in broadcast conditions.
  1. It enables development of ball-detection models in real match environments. 2. It supports evaluation of detection performance under occlusion, fast motion, and camera changes.
Yes 3.20 No China NA 2021 2021 Adult Athlete Both Professional and elite athletes across four sports (basketball, soccer, volleyball, and aerobic gymnastics) 3200 3200 3,200 video clips, 37,701 annotated action instances, 902,000 bounding boxes, from 247 competitions Cross-sectional Multiple Basketball, Soccer, Volleyball, Aerobic Gymnastics Video
  1. Action class (66 fine-grained labels). 2. Temporal boundaries (start/end frame). 3. Player bounding boxes (x, y, w, h per frame). 4. Sport type. 5. Instance ID. 6. Frame rate (25 fps). 7. Clip metadata (competition, duration).
GAN-based
WEAR WEAR: An Outdoor Sports Dataset for Wearable and Egocentric Activity Recognition Outdoor Sports Dataset for Wearable and Egocentric Activity Recognition WEAR Real Collected synchronized egocentric video (GoPro) and 3D accelerometer data from four limb-mounted smartwatches. Movement To evaluate and benchmark human activity recognition methods using wearable inertial sensors, GoPro video, and fused multimodal data in real outdoor conditions.
  1. It supports classification of outdoor activities using egocentric video. 2. It allows assessment of whether sensor–video fusion improves activity recognition accuracy.
Yes 2.85 No Germany NA 2024 2024 Adult Athlete Both Adult participants, performing outdoor workout activities 22 22 22 participants × 18 activities × 11 locations. Aprox. 19 hours of synchronized inertial and video data ( around 50 minutes per subject) Cross-sectional Fitness Fitness and outdoor bodyweight exercises (jogging, push-ups, lunges, burpees, sit-ups, stretching) Video
  1. 3D accelerometer signals (x, y, z for 4 sensors). 2. Egocentric video frames. 3. Participant metadata (ID, gender, fitness level, height, weight, frequency of workouts). 4. Session metadata (location, weather, sensor sync time). 5. Annotation timestamps (.csv, .srt).
GAN-based
SportsHHI SportsHHI: A Dataset for Human-Human Interaction Detection in Sports Videos Dataset for Human-Human Interaction Detection in Sports Videos SportsHHI Real Manual collection and annotations in basketball & volleyball videos. Player Benchmark for detection of player interactions requiring spatio-temporal reasoning.
  1. Enable development of models that detect and classify player interactions. 2. Improve multi-person video understanding in crowded sports scenes.
Yes 3.30 No China NA 2024 2024 Adult Athlete Both Professional basketball and volleyball athletes (both male and female) from international and club-level competitions 11398 11398 11,398 annotated keyframes, 50,649 human-human interaction instances, 118,075 bounding boxes across 160 sports videos (80 basketball, 80 volleyball) Cross-sectional Multiple Basketball and Volleyball Video
  1. Human bounding boxes (x, y, width, height). 2. Subject/object IDs. 3. Thirty-four interaction classes (technical, tactical, confrontational). 4. Interaction triplet ⟨S, I, O⟩. 5. Visibility flags. 6. Frame metadata. 7. Temporal continuity (5 FPS linking).
GAN-based
ScopeSense ScopeSense: An 8.5-Month Sport, Nutrition, and Lifestyle Lifelogging Dataset 8.5-Month Sport, Nutrition, and Lifestyle Lifelogging Dataset ScopeSense Real Lifelogging with digitised monitoring systems of biometrics, nutrition and training. Performance To capture sport, nutrition, wellness, and lifestyle data from two individuals for longitudinal analysis.
  1. Enable modelling of individual health, wellness, and training patterns. 2. Support the development of personalised and proactive health analytics. 3. Allow exploration of relationships between diet, activity, biometrics, and subjective well-being.
Yes 2.15 No Norway NA 2021 2021 Adult Normal Male Adult males with regular exercise and standard diet 510 2 2 participants × 255 days. Aprox. 510 participant-days; thousands of biometric, nutrition, and wellness data points Longitudinal Fitness Running, strength training, general exercise (self-logged sessions) Physiological
  1. Apple Watch: heart rate, ECG, oxygen saturation, sleep, step count, flights climbed, energy use, exercise sessions, distance, elevation. 2. Lifesum: meals, snacks, food names, brands, portion weights, nutrients (calories, carbs, fats, proteins, sodium, potassium, cholesterol), and body weight. 3. PMSys: subjective wellness (fatigue, stress, soreness, sleep duration, sleep quality, mood, readiness, injuries, training load).
Statistical
SportsMOT SportsMOT: A Large Multi-Object Tracking Dataset in Multiple Sports Scenes Large Multi-Object Tracking Dataset in Multiple Sports Scenes SportsMOT Real Manual annotation of sports videos using a customised labelling tool with propagated bounding boxes and ID tracking. Tactical To create a large-scale, sports-specific multi-object tracking dataset enabling player tracking for statistics, motion analysis, and tactical evaluation.
  1. Enable development of Multiple Object Tracking (MOT) models that handle fast and variable-speed motion in sports. 2. Support learning appearance-based association where players have similar yet distinguishable visual features. 3. Provide data for analysing player movement, team tactics, and sports performance.
Yes 3.90 No China NA 2023 2023 Adult Athlete Both Professional athletes (basketball, volleyball, football players) from real-world competition broadcasts 150379 240 240 video sequences, 150,379 frames, 1,629,490 bounding boxes, and 3,401 unique player tracks across 3 sports Cross-sectional Multiple Basketball, Volleyball, Soccer Video
  1. Bounding box coordinates (x, y, w, h). 2. Track ID. 3. Frame index. 4. Sport category. 5. Video metadata. 6. Occlusion flags.
GAN-based
BMT-Bench Bmt-Bench: A Benchmark Sports Dataset For Video Generation Badminton Benchmark BMT-Bench Real Manual collection of high-speed video recording, distortion correction, and expert labeling of badminton match clips. Tactical To provide a video benchmark enabling future frame prediction, and real-time object detection in badminton.
  1. Enable object detection and shuttlecock tracking in high-speed badminton rallies. 3. Allow fair comparison of machine learning and computer vision methods using standardised, fine-grained badminton video data.
Yes 3.30 No Taiwan NA 2023 2023 Adult Athlete Both Semi-professional badminton players from the NCU badminton team 2005 19 2,005 labeled and unlabeled video clips; 19 players; 60 FPS video recordings at 1280×960 resolution. Cross-sectional Badminton Badminton Video
  1. Player ID and side (red/blue). 2. Stroke type and sequence. 3. Score per rally. 4. Shuttlecock bounding boxes (labeled dataset). 5. Frame index and timestamps. 6. Cropped rally clips. 7. Calibration metadata for camera correction.
GAN-based
DeepSportradar-v2 DeepSportradar-v2: a Multi-Sport Computer Vision Dataset for Sport Understandings Multi-Sport Computer Vision Dataset for Sport Understandings DeepSportradar-v2 Real Computer vision Cricket datasets built from automated camera recordings. Tactical Provide benchmark datasets in cricket such as player segmentation, player re-identification, ball tracking, and bowl-release detection.
  1. Enable evaluation of computer-vision models across Cricket with different dynamics and camera conditions. 2. Support development of algorithms for segmentation, re-identification, calibration, and temporal action detection.
Yes 2.85 No International NA 2023 2023 Adult Athlete NA Professional cricket players from global matches 40 40 40 videos, 26 annotated (18 train, 8 test), 14 challenge videos Cross-sectional Cricket Cricket Video
  1. Player bounding boxes and instance masks (x, y, w, h). 2. Role labels (player, referee, coach). 3. Camera calibration matrices. 4. Re-identification image crops. 5. Cricket bowl release event tags. 6. Player and team metadata.
GAN-based
PMData PMDATA: A Sports Logging Dataset Sports Logging Dataset PMData Real Logged over five months using digitised monitoring systems biometrics, nutrition, training load, injuries and food mages from smartphone cameras. Performance To combine lifelogging and sports-activity logging into a unified dataset to enable analysis of daily life, wellness, and sports-related behaviour.
  1. Enable prediction of weight and performance changes using combined sensor and self-report data. 2. Support analysis of relationships between sleep, wellness, training load, food intake, and physiological responses.
Yes 3.40 No Norway NA 2020 2020 Adult Multiple Both Adults aged 25–60, mixed athletic backgrounds (active, former athletes, sedentary individuals). 2440 16 16 participants × 5 months = 80 participant-months; total of 2,440 activity sessions, 20,991,392 heart rate entries, 1,747 wellness reports, 225 injury logs, 1,569 daily diet reports, 644 food images. Longitudinal Fitness General physical activities running, walking, treadmill, cycling, gym workouts (self-reported) Physiological
  1. Fitbit Versa 2: heart rate, calories, steps, distance, sleep score, resting heart rate, time in HR zones, sedentary/active minutes. 2. PMSys app: training load (sRPE), fatigue, sleep duration/quality, soreness, stress, mood, readiness, injury logs (body location, severity). 3. Google Forms: daily meals, drinks, alcohol intake, weight, food frequency, fluid consumption, food images with EXIF metadata.
Statistical
C-Sports Collective Sports: A multi-task dataset for collective activity recognition Collective Sports (C-Sports) Dataset Csports Real Web-collected real sports videos. Tactical To provide a multi-sport benchmark for recognising collective activities and sports categories.
  1. Support multi-task learning for simultaneous recognition of activities and sport categories. 3. Assess how well models generalise to unseen sports contexts using the unseen-sports data.
Yes 4.15 No International NA 2020 2020 Adult Multiple Both Professional and amateur athletes from 11 sports collected from public video sources 167935 2187 2,187 video clips; 167,935 total frames; 11 sports categories × 5 collective activities Cross-sectional Multiple Football, Basketball, Dodgeball, Soccer, Handball, Hurling, Ice Hockey, Lacrosse, Rugby, Volleyball, Water Polo Video
  1. Sports category (11 total). 2. Collective activity (5 total: Gathering, Dismissal, Passing, Attack, Wandering).
GAN-based
nflfastR NA National Football League CRAN Real Scrapes and compiles NFL play-by-play data back to 1999 using JSON sources. Tactical To provide complete, and enhanced NFL game data across seasons.
  1. Allow users to analyse NFL plays with metrics such as expected points and win probability.
Yes 2.80 No US 1999 2025 1999-present Adult Athlete NA NFL players, teams 1500000 1500000 > 1.5M matches Longitudinal Football American Football Tabular Game ID, team, play, EPA, win prob Statistical
worldfootballR NA Transfermarkt, Fbref, Understat, fotmob CRAN Real Scrapes and cleans football data from FBref, Transfermarkt, Understat, and Fotmob using automated R functions. Tactical To provide access to clean and football match, player, team and league data without manual exporting from websites.
  1. Enable analysis of team and player performance across matches and seasons using standardised statistics. 2. Support extraction of detailed match events such as shots, lineups, goals, and possession for tactical or performance evaluation. 3. Facilitate multi-league comparisons using datasets from major football data providers.
No 2.95 No International NA 2024 XXXX-2024 Adult Athlete NA Soccer players, clubs 200000 200000 > 200K matches Longitudinal Soccer Soccer Tabular Match stats, shots, xG, transfers Statistical
Lahman NA Sean Lahman Baseball Database CRAN Real Compiles pitching, hitting, fielding, biographical, postseason, awards, and team statistics from 1871–2024 into a relational database structure where all tables link through unique playerID and team identifiers. Performance To provide a historical record of Major League Baseball performance, enabling efficient querying, processing, and visualisation of long-term baseball statistics.
  1. Support analysis of player and team performance trends across more than a century of baseball. 2. Enable investigation of biographical, positional, postseason, and award patterns using linked tables. 3. Allow exploration of historical league structure, franchise evolution, and cross-era comparisons using longitudinal data.
Yes 2.50 No US 1871 2024 1871-2024 Adult Athlete NA MLB players, teams 100000 100000 > 100K player Longitudinal Baseball Baseball Tabular Batting, pitching, fielding, bio Statistical
nba_api NA National Basketball Association Python Real Retrieves NBA APIs including player stats, team data, play-by-play, scoreboards, and static datasets. Tactical To enable access to NBA.com data for documentation, endpoint exploration, and real-time game information.
  1. Allow analysis of player and team performance using official NBA statistics. 2. Support retrieval of live game data such as scoreboards and game for real-time analytics. 3. Enable exploration and mapping of NBA.com endpoints to study changes, structure, and availability of league data.
Yes 2.20 No US 1996 2025 1996-present Adult Athlete NA NBA players, teams 1000 1000 > 1K matches Longitudinal Basketball Basketball Tabular Player stats, team data Statistical
pybaseball NA Major League Baseball’s Statcast system Python Real Scrapes baseball data directly from Baseball Savant, Baseball Reference, FanGraphs, Retrosheet, and the Chadwick Bureau. Tactical To provide access to MLB data including statcast events, batting/pitching metrics, standings, schedules and awards. NA Yes 2.45 No US 2015 2025 2015-present Adult Athlete NA MLB players, teams 1000000 1000000 > 1M pitches Longitudinal Baseball Baseball Tabular Pitches, outcomes, statcast metrics Statistical
cfbfastR NA C ollege football data CRAN Real Provides an R wrapper from the CollegeFootballData API to retrieve, aggregate, and tidy college football data. Tactical To enable access to college football data through the API for analysis, modeling, and performance benchmarking.
  1. Support evaluation of expected points and win probability metrics using standardised data. 2. Enable retrieval and analysis of detailed college football statistics across seasons and teams.
No 2.95 No US 2001 2025 2001-present Adult Athlete NA NCAA football players 1000 1000 > 1K matches Longitudinal Football College Football Tabular Play ID, team, EPA/WPA Statistical
wehoop NA Woman National Basketball Association CRAN Real Scrapes and aggregates women’s basketball data from the WNBA Stats API and ESPN. Tactical To allow users to retrieve and analise WNBA and women’s college basketball data through a unified R interface.
  1. Enable analysis of play sequences to study in-game events. 2. Support evaluation of team and player performance using box score and standings data. 3. Provide structured datasets for examining game outcomes and season trends across women’s basketball competitions.
Yes 2.65 No US NA 2025 XXXX-present Adult Athlete NA WNBA , college players 1000 1000 > 1K matches Longitudinal Basketball Basketball Tabular Player, team, game stats Statistical
hoopR NA National Basketball Association CRAN Real Scrapes and aggregates men’s basketball data from ESPN play and box scores, retrieves NBA Stats API data, and extracts KenPom statistics for subscribers. Tactical To provide users access to men’s basketball game, box score, team and player data across NBA and NCAA competitions.
  1. Enable analysis of live and historical play-by-play sequences with shot locations. 2. Support team and player performance evaluation using detailed box score and season data.
Yes 2.50 No US 2010 2025 2010-present Adult Athlete NA NBA players, teams 1000 1000 > 1K matches Longitudinal Basketball Basketball Tabular Boxscore, team stats Statistical
nhlapi NA National Hockey League CRAN Real Retrieves and processes data directly from the open NHL API. Tactical To provide access to NHL metadata and game-level information across multiple league entities and seasons.
  1. Allow examination of players, teams, and league structure through API metadata. 2. Enable analysis of game events, boxscores, linescores and season schedules. 3. Support exploration of drafts, prospects, awards and historical league changes using API-derived data.
Yes 2.60 No US,Canada 2010 2025 2010–present Adult Athlete NA NHL players, teams 1000 1000 > 1K matches Longitudinal Hockey Ice Hockey Tabular Player stats Statistical
cricketdata NA ESPNCricinfo and Cricsheet CRAN Real Data scraped and downloaded from ESPNCricinfo and Cricsheet. Tactical To provide international and major-competition cricket data for analysing matches, players, formats, and competitions.
  1. Allow analysis of batting, bowling and fielding performance across Tests, ODIs and T20s. 2. Enable examination of ball-by-ball events, match summaries, and competition-level structure. 3. Support investigation of player careers, metadata, and comparisons across countries and genders.
No 2.75 No Global 2000 2025 2000-present Adult Athlete NA Cricket players 50000 50000 > 50K matches Longitudinal Cricket Cricket Tabular Ball-by-ball stats Statistical
baseballr NA Baseball Reference, FanGraphs, MLB Stats CRAN Real Scrapes and downloads baseball data from online sources (Baseball-Reference, FanGraphs, MLB Stats API, BaseballSavant, Chadwick Bureau). Tactical To provide access to historical and current baseball statistics, player records, game logs, pitch-level data and team information for analytical use.
  1. Enable analysis of batter, pitcher, and team performance across custom time frames and leagues. 2. Support investigation of pitch characteristics, batted-ball events, and statcast-defined metrics such as barrels and edge classification. 3. Allow exploration of player identities, career timelines, draft records, and cross-system ID matching.
Yes 3.25 No US 2000 2025 2000-present Adult Athlete NA MLB players 1000000 1000000 > 1M matches Longitudinal Baseball Baseball Tabular Pitch, hit stats Statistical
sportsdataverse-py NA Men’s College Basketball, College Football, EPA, WPA, NFL, NHL Python Real Aggregates sports data by wrapping ESPN, nflfastR, cfbfastR, and wehoop API endpoints, enabling access to games, box scores, schedules, and expected points/win-probability metrics. Tactical To provide a Python interface for retrieving American football, basketball, and hockey data, including advanced EPA/WPA metrics and live game endpoints.
  1. Retrieve and analyse game and box scores across multiple sports. 2. Access and benchmark expected points added and win probability metrics. 3. Examine schedules, live-game data, and team or player performance using ESPN and fastR APIs.
Yes 3.10 No International NA NA XXXX Adult Athlete NA NBA, WNBA, NFL, NHL players 1000000 1000000 > 1M matches ALL Multiple Basketball, Football, Hockey, Tabular Player, team, game stats Statistical
AFL NA Australian Football League Database (AFL Tables , Footywire) Kaggle Real Manually compiled dataset created by scraping and integrating player and game statistics from afltables.com and footywire.com (2012–2024) with performance metrics and game information. Tactical To enable analysis of AFL player performance, match outcomes, and season patterns for applications such as tipping prediction and machine-learning-based player evaluation.
  1. Identify key players and performance drivers using game statistics. 2. Predict match results or tipping outcomes using historical AFL data. 3. Explore trends in player performance, environmental conditions, and match progression.
No 2.25 No Australia 2012 2024 2012-2024 Adult Athlete NA Australian Football League players 1713 1713 1713 players, 2663 games Longitudinal AFL AFL Tabular
  1. Player demographics (height, weight, DOB, position, origin); match info (date, round, venue, weather, attendance, scores). 2. Player stats (kicks, handballs, marks, goals, tackles, clangers, clearances, inside 50s, rebounds, votes, substitutions).
Statistical
EuroSoccer NA European Soccer Database (Sofifa, Football-data) Kaggle Real Scraping and merging match results, player attributes, team formations, event data, and betting odds from football-data.com and sofifa.com. Tactical To provide a database of European soccer matches, players, teams, formations, attributes, and betting information for advanced analytics and machine-learning applications.
  1. Analyse player, team, and match performance across seasons and leagues. 2. Predict match outcomes and compare model probabilities with bookmaker odds. 3. Explore tactical patterns, player attributes, and in-game events for insight generation.
No 2.85 No Belgium, England, France, Germany, Italy, Netherlands, Poland, Portugal, Scotland, Spain, Switzerland 2008 2016 2008-2016 Adult Athlete NA Professional soccer players and clubs from 11 European national leagues 10000 10000 > 25,000 matches, > 10,000 players, > 100K player attributes Longitudinal Football Football Tabular
  1. Match outcomes, scores, goals, fouls, corners, cards, possession; player attributes (FIFA-based skill ratings). 2. Team formations (X–Y coordinates). 3. Betting odds from 10 providers.
Statistical
120OlympicHistory NA 120 years of Olympic history: athletes and results (Sports-reference) Kaggle Real Scraping and wrangling athlete-level Olympic data from sports-reference.com, covering all modern Games from 1896 to 2016. Tactical To provide a historical record of athlete participation, demographics, events and medals for analysing long-term Olympic trends.
  1. Examine how Olympic participation has changed across nations, genders, and sports. 2. Assess how athlete characteristics (age, height, weight) vary across events and years. 3. Analyse medal outcomes and performance patterns across countries and time.
No 2.65 No International 1896 2016 1896-2016 Adult Athlete NA Olympic athletes (men and women) representing all participating countries across Summer and Winter Games 271116 271116 271116 athlete-event records; 15 columns Longitudinal Olympic Olympic sports Tabular ID, Name, Sex, Age, Height, Weight, Team, NOC, Games, Year, Season, City, Sport, Event, Medal Statistical
FIFAWC NA FIFA World Cup Dataset Kaggle Real Compiling historical World Cup and match-level results from the official FIFA World Cup Archive. Tactical To provide a historical record of tournaments and match outcomes for analysis, modelling, and comparison across World Cup.
  1. Analyse historical tournament trends and team performance across World Cups. 2. Study match outcomes and identify patterns over different eras. 3. Build predictive models to estimate future World Cup winners.
No 2.15 No International 1930 2014 1930-2014 Adult Athlete NA National football teams, players, referees, and tournament-level outcomes across all FIFA World Cups 7633 7633 > 7633 records Longitudinal Football Football Tabular
  1. Tournament (year, host country, winner, runner-up, third, fourth, goals, attendance, matches played). 2. Match-level (teams, stadium, city, goals, attendance, referee, stage, half-time goals, win conditions). 3. Player-level (team initials, coach, lineup, shirt number, player name, position, event: goals, yellow/red cards, substitutions).
Statistical
NHL NA NHL Game Data Kaggle Real Extracting all official NHL game metrics and detailed game events from the undocumented NHL stats API over the past six years. Tactical To build a game and player data to enable hockey analytics.
  1. Identify which individual on-ice actions contribute most to game outcomes. 2. Build predictive models to improve winner prediction accuracy in NHL games.
No 2.95 No Canada, US 2013 2018 ~ 2013-2018 Adult Athlete NA Professional ice hockey players, teams, officials, and game events across multiple NHL seasons 3000000 3000000 > 3M records, players, stats, games, event Longitudinal Hockey Ice Hockey Tabular
  1. Game info (date, teams, goals, rink side, venue, time zone). 2. Player stats (goals, assists, shots, hits, blocks, penalties, time on ice). 3. Goalie stats (saves, shots, goals against, save %). 4. Play data (x,y coordinates, event type, time, score). 5. Team stats (faceoffs, giveaways, takeaways, power plays). 6. Metadata (officials, scratches, shifts, penalties).
Statistical
NFBDB2026 NA NFL Big Data Bowl 2026: Player Tracking and Movement Prediction Kaggle Real Tracking dataset created from NFL Next Gen Stats player-tracking data. Player To predict player movement after the pass is thrown during NFL plays.
  1. Predict player x–y positions for future frames in a passing play. 2. Understand how player speed, acceleration, orientation, and role influence movement outcomes. 3. Build forecasting models usable on unseen live-game tracking data in competition scoring.
Yes 2.35 No US NA 2023 2023 Adult Athlete NA Professional NFL players 18 18 18 weeks of game data Longitudinal Football American Football Tabular Game ID, play ID, frame ID, nfl_id, player name, position, height, weight, direction, speed (s), acceleration (a), orientation (o), play direction, yardline, ball landing coordinates (x, y) Statistical