NBA and International Basketball Analytics Portfolio
MOLINAR NBA Draft Projection Model
My first ever basketball analytics project, MOLINAR attempts to project the probability of a player from either college or international basketball getting selected in the NBA Draft. It uses an XGBoost framework with additional weights for strength of conference, position, class/age, and performance relative to others.
Machine Learning NBA Draft Big Board
Using over 500 variables from a multitude of sources, I built a family of machine learning models that attempt to project players' performances over the course of their rookie contracts. These projections are then rank ordered to create a final "big board" entirely based on who the model thinks will perform well. The rankings are hosted on the Tawny Park Metrics website (click on the picture for an article explaining my methodology).
Evaluating Player Spacing Using Computer Vision
My attempt to use computer vision techniques, notably YOLO and DeepSORT, to evaluate and grade player spacing using only broadcast data as an input. Wrote about my process in a longform article.
Context-Adjusted Points Over Expected
A novel statistic which ranks every shot taken in the NBA by difficulty (shot distance, type, player height, etc.) opponent strength, and context (time of game, point differential, importance of shot). The shots with the highest CPOE are therefore the most difficult and contextually important shots of a given season, and vice versa.
International and FIBA Box Plus-Minus Database
The most comprehensive advanced statistics database that is publicly available, my collection includes BPM computations for over 75 domestic and continental competitions, as well as every FIBA Youth (U17 and above) and ANGT event. This tool has become extremely important for scouts and front office personnel at all levels.
The Combine Score database and interactive Python dash webapp grades the athleticism of every player who has participated in the NBA Draft Combine since 2000 on a scale from 0 to 100, with a higher score indicative of a player being more of an overall athlete compared to those at his position historically.