This project aims to predict the statistical performance of college basketball players upon entering the NBA by integrating structured data (rookie-year statistics) with unstructured data (text-based scouting reports). The proposed system applies NLP techniques to extract features from scouting narratives, focusing on player strengths, weaknesses, and comparisons. These textual features are combined with quantitative performance metrics to develop predictive models for key statistical categories such as points, assists, rebounds, blocks, and steals.
Clustering methods are further applied to group players with similar scouting profiles. Instead of labeling clusters as fixed archetypes, the approach emphasizes identifying the most comparable NBA players for each prospect. This similarity-based framework provides interpretability for scouting evaluations and a practical foundation for projecting performance. The predictive component then leverages historical outcomes of comparable players to generate forecasts for new prospects.


The results demonstrate that clustering can organize players into meaningful groups based on textual analysis, while the predictive framework produces reasonable estimates of rookie-year performance. Together, these findings highlight the potential of integrating NLP and ML into scouting workflows to complement traditional evaluation methods, reduce bias, and enhance the accuracy of performance projections.
]]>