NBA Win Prediction Engine

Predictive Modeling for the 2025-26 Season

Created by Timur Yesmukhan | Data Analytics Portfolio

The Objective

How well do Advanced Efficiency Metrics predict regular-season success in the modern NBA?

  • Extract 5 years of historical team data
  • Analyze the correlation between Pace and Net Rating
  • Build a predictive model for the current season

Tech Stack & Pipeline

Extraction: Python & nba_api

Modeling: Scikit-Learn (Linear Regression)

Frontend: Streamlit & Plotly

Storage: Pandas Dataframes / CSV

Feature Engineering

I focused on four key "Pillars" of team performance:

Net Rating
Pace
Off. Efficiency
Def. Efficiency

2025-26 Results

The model predicted OKC Thunder (63.8 vs 64 wins) with near-perfect accuracy.

Mean Absolute Error (MAE): ~2.8 Wins

Identifying Anomalies

Why did the model miss on the Lakers?

  • Prediction: 45 Wins | Actual: 53 Wins
  • Insight: Statistical "Overperformance" in clutch minutes suggests factors beyond raw efficiency are at play (Veteran IQ/Coaching).

The Dashboard

I deployed a live Streamlit app for real-time "What-If" analysis.

Explore the Live Dashboard