Revisiting the Correlation of Basketball Stats and Match Outcome Prediction

Abstract

Professional basketball games, high-stake in nature, have garnered great attention from the research community to model, simulate and predict game outcomes. Despite all the advanced metrics and models developed over the years, predictions remain unreliable and upsets still occur frequently. This paper reinvestigates whether these complex stats perform significantly better over simple statistics, and whether simple statistics benefit from common machine learning generalization and robustification processes.