Summary: “The NFL Big Data Bowl is an annual, global crowdsourcing competition where analysts, students, and professionals use the NFL's Next Gen Stats player tracking data to generate novel and actionable statistics to enhance the game of football. Participants create "Notebook" submissions that are evaluated for their innovation, insights, and potential impact on strategy for NFL teams. The competition is powered by AWS and culminates with finalists presenting their work to NFL teams at the Scouting Combine.”

Summary: A team of 5-6 SAGB analysts will be joining a case competition later this semester. Last year, SAGB presented a study defining the "middle-class" of NFL players and used it to present recommendations.

Summary: Two teams of SAGB analysts will work on independent mini-research labs: each team will pick a focused baseball analytics question, review past work, and build models/analysis to generate new insights. One team will tackle a general baseball performance/economics problem (e.g., player evaluation, roster strategy), while the other will explore a biomechanics angle (e.g., athlete load tracking with Catapult data). By November 21, both teams will synthesize their methods, preliminary results, and expected contributions into a polished abstract submission.

Summary: The Minnesota Lynx are interested in targeting overseas players in this upcoming free agency. The staff needs to predict international players’ success when crossing over to the WNBA. This project focused on projecting international players' success in the WNBA, as well as finding international player comparisions with current WNBA players.

Summary: In a collaboration between the Business and Projects Divisions, Sports Analytics Group at Berkeley (SAGB) will be responsible for assisting Athvest in their athlete valuation model, creating a framework for how Athvest will present their athlete values to both student-athletes and investors, along with aiding their funding and partnership efforts in any way possible.

Summary: In collaboration with Sports Aptitude, SAGB developed a Flask app used to generate different scouting reports for college basketball teams. We focused on storing shot quality and shot tendency data for NCAA athletes into a database, as well as developing the front-end user interface with Python and HTML.

Summary: This project focused on visualizing tracking data of swings for minor league players in the San Francisco Giants Minor League affiliates to help scouts and the Giants front office better evaluate players.

Abstract: This research creates a classification-based model that predicts the salary amounts given to NBA players in free agency. This was done through collecting player salary data through Spotrac and player statistics through Basketball Reference. Correlation analysis demonstrated that points, turnovers, VORP, and rebounds created a balanced and effective set of variables for unweighted KNN-classification. The former three statistics were standardized by year, while the latter statistic was standardized by year and position due to underlying relationships between those categories. After classification, the model revealed that volume statistics were most indicative of salaries within lower and mid-level players, while high VORP was common among all players making near-max to max salaries. Using test set data, the model was off by 1.21 clusters on average, but many outliers were due to gross over and undervaluation on the part of the franchises. The model was used to predict salaries for incoming free agents in the 2018 class, with reasonable projections that meet qualitative expectations. Click for the full report!

Summary: This project focused on scraping data from the Denver Nuggets Twitter to perform sentiment analysis and measure current sentiment towards the team and players. A major aspect of this project was finding trending keywords and names during the regular season.

Abstract: We present a shot classifier for the NBA trained on positional player data and player ratings data that classifies a shot attempt into a make or a miss. This classifier has an accuracy of 56.23% on test data, and the precision, recall, and f1­-score of our classifier was 0.54, 0.56, and 0.51 respectively. We also present a shot evaluator as an extension of this classifier that outputs a value, 0 to 1, corresponding to the efficiency of a shot. The evaluator was created using a combination of the probability values of our classifier and the type of field goal attempt. Click for the full report!

Summary: This project focused on using machine learning to project individual player performance 5 years into the future.

Abstract: This research creates and offers a new metric called Free Agency Rating (FAR) that evaluates and compares franchises in the National Football League (NFL). FAR is a measure of how good a franchise is at signing unrestricted free agents relative to their talent level in the offseason. This is done by collecting and combining free agent and salary information from Spotrac and player overall ratings from Madden over the last six years. This research allowed us to validate our assumptions about which franchises are better in free agency, as we could compare them side to side. It also helped us understand whether certain factors that we assumed drove free agent success actually do. In the future, this research will help teams develop better strategies and help fans and analysts better project where free agents might go. The results of this research show that the single most significant factor of free agency success is a franchise’s winning culture. Meanwhile, factors like market size and weather, do not correlate to a significant degree with FAR, like we might anticipate. Click for the full report!