Data Mining Computing of Predicting NBA 2019–2020 Regular Season MVP Winner

摘要

This project is to build statistical models to decide which players should win the NBA 2019–2020 regular season Most Valuable Player (MVP) Award. Top 50 potential candidates were selected to enter the 2019–2020 MVP race. Prior building the MVP models, the player statistics data has been Z-standardized to remove any mean and standard deviation bias. The “Uniform MVP Index” has been derived from combining each player's Z statistics with equal weight. Team has further derived a “Weighted and Subset” model by adding the weight factor and the best subset feature selection. Authors have added the “Team Winning” factor in the Power Model from power= 0 (equivalent to the Weighted Model), 0.5,1,1.5 to power= infinity (MVP from the best Team). The Power MVP Index will be multiplied by the power of the team winning% in the Power model. 9 different MVP index were established based on top 50 selected players' statistics and team records. Modeling overfit risk was addressed by multivariate correlation, and recursively partition. Neural algorithms were utilized to build the MVP prediction model based on MVP Index methods.