Decoding the Draft: Enhancing NBA Prospect Predictions with Quantified Scouting

Abstract

The world of sports has advanced tremendously over the last decade, utilizing analytics and technology to gain strategic advantages. It is not enough to have the best strategy, the best players, or the best facilities, and sports analytics has seen phenomenal growth and interest to help teams gain any competitive edge. This focus on advanced metrics has been especially evident in professional basketball, where teams invest in data to supplement traditional strategy. Advancements in technology and tracking player metrics have improved how teams operate in every facet of the game, from free-agency signings, mid-season trading, salary negotiations, competitive strategy, player health and management, player development, and many more. However, the one piece of basketball that seems unchanged is player evaluation. Teams have clung to traditional methods of player evaluation, heavily favoring subjective scouting reports and "eye tests." Quantitative analysis has progressed past the standard box score statistics---like points, rebounds, and assists---and developed various advanced metrics to understand player performance on a deeper level. Qualitative analysis remains subjective and immeasurable, offering a potential opportunity to understand player performance better. This work offers a pathway for comprehensively evaluating an NBA draft prospect, considering quantitative and qualitative factors. By training an a large language model (LLM) to conduct sentiment analysis on qualitative scouting reports and quantify that data, this thesis illustrates a method for assessing a prospect's career performance by merging two previously separated aspects. Through this work, NBA teams could build upon this foundation to select the best player available during the draft process, backed by quantitative and qualitative analysis.