NBA Player Position Prediction using K-Nearest Neighbors (KNN)

Project Overview

This project implements a machine learning pipeline to classify NBA players' positions based on their performance statistics using the K-Nearest Neighbors (KNN) algorithm. It involves data preprocessing, model training, validation, and evaluation, including detailed metrics and insights.

---

Features

• Exploratory Data Analysis (EDA): Ensures no null or zero values in the dataset.
• Label Encoding: Encodes categorical data (positions) into numerical labels for modeling.
• Model Training and Validation: Splits the dataset into training and validation subsets.
• Performance Evaluation:
  • Calculates training, validation, and testing accuracy.
  • Produces confusion matrices for in-depth error analysis.
  • Computes position-wise classification accuracy.
• Cross-Validation: Employs 10-fold stratified cross-validation for robust evaluation.

---

Key Results

Accuracy

• Training Accuracy: 63.60%
• Validation Accuracy: 61.40%
• Testing Accuracy on Dummy Dataset: 62.14%

Confusion Matrices

Training Confusion Matrix:

[[106  18   0   5   0]
 [ 27  79   2  19  11]
 [  4   8 100   2  20]
 [  9  29   6  64  23]
 [  3  15  23  25  86]]

Validation Confusion Matrix:

[[23  4  0  2  0]
 [11 16  1  7  6]
 [ 0  0 28  0  4]
 [ 0  6  2 13  8]
 [ 0  4  8  3 25]]

Testing Confusion Matrix:

[[15  1  0  0  0]
 [ 5 10  0  1  4]
 [ 0  0 17  0  1]
 [ 1  5  2  9 10]
 [ 1  3  1  4 13]]

Cross-Validation

• Scores per Fold: [0.4419, 0.5116, 0.5698, 0.5000, 0.5000, 0.5059, 0.5529, 0.5294, 0.6118, 0.4824]
• Mean Cross-Validation Accuracy: 52.06%

Position-Wise Classification Accuracy on Test Set:

• C: 93.75%
• PF: 50.00%
• PG: 94.44%
• SF: 33.33%
• SG: 59.09%

---

Dataset

The project uses the following datasets:

1. Training Dataset: nba_stats.csv
2. Testing Dataset: dummy_test.csv

Features Used:

• AST, STL, BLK, TRB, FGA, FG%, 3P, 3PA, PF, eFG%, FT%, DRB, 3P%, 2P, 2P%

---

Prerequisites

• Python (3.x)
• Required Libraries:
  • pandas
  • scikit-learn

Installation

Install the dependencies using pip:

pip install pandas scikit-learn

---

Usage

1. Dataset Preparation: Place nba_stats.csv and dummy_test.csv in the working directory.
2. Run the Script: Execute the NBA_KNN.py file:

   python NBA_KNN.py

3. Outputs:
   • Accuracy scores
   • Confusion matrices
   • Cross-validation results
   • Position-wise classification accuracies

---

References

• K-Nearest Neighbors Algorithm in Python
• Scikit-learn Documentation

---