NHL Player Performance Prediction

by Team Lewandowski

(Elvin Kim, Jay Leung, Adian Murzagaliyev, Sou Hamura)

Objective

This project aims to develop a projection system that predicts a hockey player’s performance during their first three NHL seasons based on their statistics in previous leagues. The system processes an original dataset, automates data retrieval from HockeyDB, and applies machine learning to predict future performance.

Why This Project Matters & What Makes It Unique

This project is deeply connected to who we are—as members of the UCSB hockey team, we are combining our passion for hockey with data science to solve a longstanding challenge in player scouting.

Why Predicting NHL Performance Is Difficult

Unlike the NFL or NBA, where most players come from a single college system (NCAA), NHL players come from dozens of different leagues across multiple countries, each with varying levels of difficulty.

✅ Different Leagues, Different Standards – Scoring in the KHL, SHL, AHL, or NCAA is not the same.
✅ Same Country, Multiple Leagues – In North America, players can develop in the NCAA, OHL, WHL, or USHL, all with different competition levels.
✅ International Complexity – Players from Europe, Russia, and North America follow unique development paths, making performance comparisons difficult.

What Makes Our Model Unique?

Our Random Forest-based model overcomes these challenges by:
✅ Adjusting for League Difficulty – Captures how hard it is to produce points in different leagues.
✅ Learning from Global Data – Incorporates stats from players across all major pre-NHL leagues.
✅ Built by Players, for Players – As hockey players, we understand the nuances of player development beyond just numbers.