Improving upon NBA point-differential rankings

Abstract

For some time, point-differential has been thought to be a better predictor for future NBA success than pure win-loss record. Most ranking and team performance predictions rely largely on point-differential, often with some normalizations built-in. In this work, various capping and weighting functions are proposed to further improve indicator performance. A gradient descent algorithm is also employed to discover the optimized weighting/capping function applied to individual game scores throughout the season.