How Does This Work? - NBA Player Salary Prediction

Overview

This project uses supervised machine learning (linear regression) to predict NBA player salaries based on their statistics. The model compares predicted salaries with actual salaries to determine if players are overpaid or underpaid.

Data Collection

We collect data about NBA players using the SportsData.io API, which provides comprehensive statistics and salary information for over 340 players. This data is organized and stored in a SQLite database for efficient access.

Feature Selection

We analyzed over 50 different statistics to identify the most relevant features for salary prediction. Using Pearson correlation coefficients, we selected 28 parameters that have the strongest relationship with player salaries.

Points vs Salary Correlation — Points vs Salary (Correlation: 0.73)

3-Point Percentage vs Salary Correlation — 3-Point % vs Salary (Correlation: 0.05)

Model Training

Our model uses linear regression with gradient descent optimization. The process includes:

Z-score normalization of features to ensure consistent scaling
Mean squared error as the cost function
Gradient descent optimization over 100,000 iterations

Cost Function Formula — Mean Squared Error Cost Function

Gradient Formula — Gradient Descent Update Rule

Prediction Process

When you enter a player's name:

The system fetches the player's statistics from our database
Features are normalized using pre-computed means and standard deviations
The model predicts the player's expected salary
Results are compared with actual salary to determine if the player is overpaid or underpaid

Technical Stack

This project is built using:

Python with Flask for the backend
SQLite for data storage
NumPy for numerical computations
Matplotlib for data visualization
Modern HTML5, CSS3, and JavaScript for the frontend