--- title: "20 Best Free Sports Datasets for ML 2025" description: "Sports data is your playbook: choose right, win fast. This multi-sport, ML-ready shortlist includes free + paid options, a quick comparison matrix, and clear notes…" url: "https://unidata.pro/blog/best-free-sports-datasets-ml/" date_modified: "2026-05-28T13:00:49+03:00" language: "en-US" --- ![20 Best Free Sports Datasets for ML 2025](https://unidata.pro/wp-content/uploads/2025/10/sports-dataset.webp)Sports data is your playbook: choose right, win fast. This multi-sport, ML-ready shortlist includes free + paid options, a quick comparison matrix, and clear notes on how to plug each dataset into live pipelines (prediction, CV, tracking). Open Football APIs for Real-Time Modeling ----------------------------------------- ### **1.**[ StatsBomb Open Data](https://github.com/statsbomb/open-data) ![StatsBomb Open Data](https://unidata.pro/wp-content/uploads/2025/10/statsbomb-scaled.webp)**Volume:** 30+ competitions, thousands of matches since 2018 **Access:** Free with attribution **Format:** JSON event files + CSV match data **Task Fit:** Event classification, player analysis, match prediction If “context is king,” this is royalty—pressures, pass heights, shot freeze-frames, the lot. It punishes lazy features and rewards smart ones (xG, buildup chains, pitch zones). If your model reads the game instead of raw rows, it’ll shine here. ### **2.** [Open Football Data API](https://www.football-data.org/) ![Open Football Data API](https://unidata.pro/wp-content/uploads/2025/10/open-football-scaled.webp)**Volume:** Live & historical match results, fixtures, odds **Access:** Free API (registration required) **Format:** REST API with JSON responses **Task Fit:** Predictive modeling, betting analytics, match classification Plug-and-play football feeds without the plumbing drama. Great for spinning up live win-probability, odds-driven features, and alerting dashboards. Mind the rate limits, cache smartly, and your models stay real-time sharp. ### **3.** [College Football Data API](https://collegefootballdata.com/) ![College Football Data API](https://unidata.pro/wp-content/uploads/2025/10/college-football-scaled.webp)**Volume:** 1,000+ games per season **Access:** Free API **Format:** REST API with JSON (games, drives, plays, rosters) **Task Fit:** Win prediction, recruitment analysis, player performance Saturday chaos, structured. Play-by-play, rosters, and drive data let you model tempo, field position, and coaching tendencies. If your features capture scheme and pace, expect serious lift on win-probability curves. Player Tracking Datasets (Basketball & Football) ---------------------------------------------------- ### **4.** [Kaggle: NBA Shot Logs (2014–15)](https://www.kaggle.com/datasets/dansbecker/nba-shot-logs) ![Kaggle: NBA Shot Logs (2014–15)](https://unidata.pro/wp-content/uploads/2025/10/kaggle-nba-shot-logs-scaled.webp)**Volume:** 128,000+ shots from 2014–15 NBA season **Access:** Free (Kaggle account required) **Format:** CSV (shot location, outcome, context) **Task Fit:** Shot prediction, spatial analysis, player efficiency A slam-dunk playground for spatial models: shot location, outcome, defender context. Perfect for heatmaps, shot quality, and player profiles without chasing proprietary feeds. If distance and angle make it into your features, buckets follow. ### **5.** [SoccerNet ](https://www.soccer-net.org/) ![SoccerNet](https://unidata.pro/wp-content/uploads/2025/10/soccernet-scaled.webp)**Volume:** 500+ full broadcast matches with event labels **Access:** Free (research registration required) **Format:** Video frames, bounding boxes, JSON event annotations **Task Fit:** Player tracking, action recognition, event detection The gold standard for football video ML. Synchronized multi-camera footage, broadcast commentary, and precise event tags make it ideal for benchmarking. If your detector can survive motion blur and crowd noise here, it’s ready for prime time. ### **6.**[ Metrica Sports Sample Data](https://github.com/metrica-sports/sample-data) ![Metrica Sports Sample Data](https://unidata.pro/wp-content/uploads/2025/10/metrica-sports-scaled.webp)**Volume:** Full-match tracking + event logs **Access:** Free (GitHub) **Format:** CSV/JSON tracking coordinates + synchronized event data **Task Fit:** Player tracking, tactical analysis, computer vision Think of it as GPS for 22 dots sprinting, passing, and colliding. You get both event logs and full-match positional streams, perfectly synced. A sandbox for anyone testing CV models or tactical visualizations beyond static stats. Historical Box Scores for Outcome Prediction -------------------------------------------- ### **7.** [Sports Reference](https://www.sports-reference.com/) ![Sports Reference ](https://unidata.pro/wp-content/uploads/2025/10/sports-reference-scaled.webp)**Volume:** Decades of MLB, NBA, NFL, NHL data **Access:** Free **Format:** Web tables + downloadable CSVs **Task Fit:** Trend analysis, win prediction, player projections The encyclopedia every U.S. sports analyst secretly bookmarks. Box scores, advanced stats, and historical leaders make it prime territory for long-range forecasting. If your model can’t find signal here, it probably won’t find it anywhere. ### **8.** [Lahman Baseball Database](https://sabr.org/lahman-database/) ![Lahman Baseball Database](https://unidata.pro/wp-content/uploads/2025/10/lahman-baseball-scaled.webp)**Volume:** Over 150 years of MLB stats **Access:** Free download **Format:** CSV/SQL database files **Task Fit:** Historical trend analysis, performance prediction Baseball’s memory palace, digitized. From dead-ball era oddities to modern OPS+, it’s all in structured tables. A dream dataset for time-series experiments that span generations of players and shifting styles of play. ### **9.**[ Division III Basketball Play-by-Play](https://www.kaggle.com/datasets/samcd95/division-3-basketball-game-play-by-play) ![Division III Basketball Play-by-Play](https://unidata.pro/wp-content/uploads/2025/10/division-scaled.webp)**Volume:** 300,000+ plays from multiple Division III games **Access:** Free (Kaggle) **Format:** CSV logs with timestamps, players, and events **Task Fit:** Sequence modeling, outcome prediction, time-series A raw look into small-college basketball where structure meets chaos. Every pass, foul, and run of play is timestamped—perfect for training models that understand momentum and clutch shifts. Ideal for testing RNNs, LSTMs, or transformers built for sports flow. ### **10.** [NHL Play-by-Play Data](https://www.kaggle.com/datasets/s903124/nhl-playbyplay-data-from-2007) ![NHL Play-by-Play Data](https://unidata.pro/wp-content/uploads/2025/10/nhl-play-by-play-data-scaled.webp)**Volume:** 10+ years of NHL logs **Access:** Free (Kaggle) **Format:** CSV event logs with shots, penalties, goals **Task Fit:** Shot analysis, win prediction, efficiency metrics Hockey isn’t chaos—it’s structured chaos, and this dataset proves it. Play-by-play sequences let you analyze shot quality, penalty impact, and even goalie hot streaks. A sturdy launchpad for predictive hockey analytics. Event-Level Sports Data for xG & Tactics -------------------------------------------- ### **11.** [**FIFA 23 Player Dataset**](https://www.kaggle.com/datasets/stefanoleone992/fifa-23-complete-player-dataset?utm_source=chatgpt.com) ![FIFA 23 Player Dataset ](https://unidata.pro/wp-content/uploads/2025/10/fifa-23-scaled.webp)**Volume:** 19,000+ players, 100+ attributes **Access:** Free (Kaggle) **Format:** CSV (player attributes, positions, clubs, nations) **Task Fit:** Classification, clustering, scouting Ratings, traits, and roles—enough signal to build a scouting engine that actually feels smart. Slice by league, position group, or age curve and surface “hidden gems” your rivals overlook. Great playground for similarity search, role archetyping, and squad planning. ### **12.**[ Football Manager Complete Dataset](https://www.kaggle.com/datasets/furkanuluta/football-manager-22-complete-player-dataset?utm_source=chatgpt.com) ![Football Manager Complete Dataset](https://unidata.pro/wp-content/uploads/2025/10/football-manager-scaled.webp)**Volume:** 150,000+ players **Access:** Free (Kaggle) **Format:** CSV (player stats, attributes, positions, nations) **Task Fit:** Recommendation, scouting analysis A cult dataset reborn — clean, deep, and refreshingly current. Attribute-rich player profiles make it perfect for training recommender systems or similarity searches. Whether you’re matching midfield archetypes or ranking potential signings, this one’s pure transfer gold. ### **13.** [WTA & ATP Tennis Stats and Results](https://www.kaggle.com/datasets/taylorbrownlow/atpwta-tennis-data) ![WTA & ATP Tennis Stats and Results](https://unidata.pro/wp-content/uploads/2025/10/wtaatp-tennis-scaled.webp)**Volume:** WTA and ATP matches from 1949–2021 **Access:** Free (Kaggle) **Format:** CSV (match results, player stats, tournament metadata) **Task Fit:** Outcome prediction, ranking models Seven decades of tennis history — Grand Slams, upsets, and dominance cycles captured in one dataset. Ideal for modeling Elo-style ratings, predicting match outcomes, or studying era-based performance trends. If your model respects surface and fatigue, this set rewards nuance. Multi-Sport APIs and Data Sources --------------------------------- ### **14.** [balldontlie NBA API](https://www.balldontlie.io/) ![balldontlie NBA API ](https://unidata.pro/wp-content/uploads/2025/10/balldontlie-nba-api-scaled.webp)**Volume:** Historical & current NBA games, players, and stats **Access:** Free (no key required; sensible rate limits) **Format:** REST API with JSON responses **Task Fit:** Real-time dashboards, trend analysis, prediction features Clean, consistent NBA endpoints without scraping drama. Pull games, box scores, players, and season splits straight into notebooks or BI tools. Great for building live tiles, baseline models, and stat pipelines in a single afternoon. ### **15.**[ Sports Stats API](https://sports-statistics.com/sports-data/sports-data-sets-for-data-modeling-visualization-predictions-machine-learning/?utm_source=chatgpt.com) ![Sports Stats API](https://unidata.pro/wp-content/uploads/2025/10/sports-stats-api-scaled.webp)**Volume:** Covers football, basketball, hockey, tennis **Access:** Free tier + paid plans **Format:** REST API with JSON (multi-sport endpoints) **Task Fit:** Multi-sport modeling, visualization, predictions One doorway, many sports. Pull consistent JSON across leagues, wire it into your ETL, and ship a unified analytics layer fast. Ideal for teams that need breadth without juggling five different vendor schemas. ### **16.**[ ESPN Sports Data via Flipside LiveQuery](https://flipsidecrypto.xyz/livequery/espn) ![ESPN Sports Data via Flipside LiveQuery ](https://unidata.pro/wp-content/uploads/2025/10/espn-1-scaled.webp)**Volume:** Scores, schedules, and player stats across major U.S. sports **Access:** Free (requires Flipside account) **Format:** SQL-based API queries returning JSON/CSV **Task Fit:** Trend analysis, visualization, performance tracking Finally—ESPN data without the scraping pain. Query real game stats, schedules, and leaderboards directly through SQL endpoints. Ideal for analysts who want clean pipelines from ESPN’s ecosystem into BI dashboards or ML notebooks in minutes. ### **17.**[ FiveThirtyEight Sports Data](https://github.com/fivethirtyeight/data) ![FiveThirtyEight Sports Data](https://unidata.pro/wp-content/uploads/2025/10/fivethirtyeight-sports.avif)**Volume:** Multiple datasets (NBA, NFL, MLB, more) **Access:** Free (GitHub) **Format:** CSV with documentation/READMEs **Task Fit:** Prediction, sports betting, storytelling The datasets behind headline-grabbing forecasts, packaged for immediate use. Clean columns, sensible dictionaries, and repeatable structures make baselines quick to build. Great for demos, benchmarks, and explainable models your PM can love. ### **18.**[ DataHub Football Data Collection](https://datahub.io/blog/football-data-our-open-source-collection-of-worldwide-statistics) ![DataHub Football Data Collection](https://unidata.pro/wp-content/uploads/2025/10/datahub-football-scaled.webp)**Volume:** 60K+ match results from global leagues and tournaments **Access:** Free (open source, downloadable CSV/JSON) **Format:** CSV/JSON (team stats, results, goals, standings) **Task Fit:** Experimental modeling, benchmarking, reproducibility A clean, structured, and open dataset that brings worldwide football stats to your fingertips. No scraping, no rate limits—just tidy data ready for ML models, dashboards, or quick EDA. Ideal for testing match outcome prediction or transfer learning across leagues. ### **19.** [Match Charting Project – Tennis Data](https://github.com/JeffSackmann/tennis_atp) ![Match Charting Project – Tennis Data ](https://unidata.pro/wp-content/uploads/2025/10/match-charting-project.avif)**Volume:** 60K+ ATP & WTA matches (1968–2024) **Access:** Free (open GitHub repo) **Format:** CSV (match results, players, stats) **Task Fit:** Outcome prediction, ranking models, time-series A long-running open tennis dataset curated by Jeff Sackmann. Clean, consistent columns for player, surface, round, and result — perfect for building predictive models or ranking algorithms without any preprocessing. ### **20.** [UCI Sports Datasets](https://archive.ics.uci.edu/dataset/256/daily+and+sports+activities) ![UCI Sports Datasets](https://unidata.pro/wp-content/uploads/2025/10/uci-sports-scaled.webp)**Volume:** Small-to-mid datasets (athletics, gym, swimming) **Access:** Free **Format:** CSV/ARFF; some sensor streams **Task Fit:** Classification, biomechanics, activity recognition A classic playground for quick experiments and teaching notebooks. Sensor-rich tasks like activity recognition let you test pipelines without heavy ETL. When you need clean, compact data to prove a point, start here. **21.** [**Sports-1M**](https://cs.stanford.edu/people/karpathy/deepvideo/) ![](https://unidata.pro/wp-content/uploads/2025/10/sports-1m-1024x379.webp)**Volume:** 1,133,158 YouTube video clips across 487 sport categories **Access:** Free (open access, Google Research) **Format:** YouTube video URLs with class labels (CSV) **Task Fit:** Action recognition, video classification pretraining, sport-type detection The ImageNet of sports video. 1M+ clips, 487 classes — too large to fine-tune from scratch, perfect for pretraining backbone models before you go narrower. If you're building any sports video CV pipeline, odds are your transfer weights trace back here. Scale first, specialize second. Conclusion ---------- From detailed football event logs to real-time APIs spanning dozens of sports, these datasets cover the full spectrum of analytics needs. Whether you’re modeling match outcomes, building scouting engines, or training CV models, there’s a dataset here to fuel your project. ## Additional Modules ### FAQ Section **Questions and Answers:** - **Question:** What is a sports dataset and how is it used in machine learning? — **Answer:** A sports dataset is a structured collection of data related to games, players, teams, or events, such as match results, player statistics, tracking data, or video annotations. In machine learning, a sports dataset is used to build models for tasks like match outcome prediction, player performance analysis, scouting, and real-time analytics. These datasets power applications in sports betting, coaching strategy, and fan engagement. - **Question:** What are the best types of sports datasets for predictive modeling? — **Answer:** The best sports dataset for predictive modeling depends on your goal. Event-level datasets (like play-by-play logs) are ideal for match predictions, while player tracking datasets are better for spatial analysis and computer vision tasks. Historical box score datasets work well for long-term trend analysis, and real-time API-based sports datasets are essential for live predictions and dynamic dashboards. - **Question:** How do I choose the right sports dataset for my project? — **Answer:** To choose the right sports dataset, consider the data format (CSV, JSON, API), level of detail (event-level vs. summary stats), and whether you need real-time or historical data. You should also match the dataset to your task, such as classification, prediction, or tracking. High-quality labeling, consistent structure, and domain relevance are key factors for building reliable machine learning models. - **Question:** Are free sports datasets good enough for machine learning projects? — **Answer:** Yes, many free sports datasets are high quality and suitable for machine learning, especially for prototyping, research, and benchmarking. Open datasets like football event logs, basketball shot data, and tennis match records provide rich features for modeling. However, for production systems or advanced analytics, paid sports datasets may offer more depth, coverage, and real-time access. - **Question:** What features should I look for in a high-quality sports dataset? — **Answer:** A high-quality sports dataset should include accurate and well-labeled data, consistent formatting, and sufficient volume for training models. Important features may include timestamps, player positions, event types, and contextual information like match conditions. For advanced use cases, look for datasets with tracking data, video annotations, or real-time updates to improve model performance and reliability. [Full list of this site's AI-readable pages](https://unidata.pro/llms.txt)