Decision Tree from Scratch

This project implements a Decision Tree Classifier from scratch using only NumPy, Pandas, and Matplotlib, without using Scikit-Learn. The dataset used is the Banknote Authentication Dataset from Kaggle.

📌 Features

• Pure NumPy Implementation (No Scikit-Learn)
• Custom Gini Impurity Calculation
• Manual Data Shuffling & Splitting
• Recursively Built Decision Tree
• Performance Evaluation (Accuracy Calculation)

📂 Dataset

The dataset contains 4 numerical features extracted from banknotes:

• Variance of Wavelet Transformed Image
• Skewness of Wavelet Transformed Image
• Curtosis of Wavelet Transformed Image
• Entropy of Image

The target column:

• Class (0: Fake Note, 1: Authentic Note)

🔧 Installation

Clone this repository and install dependencies:

https://github.com/KartikAg13/banknote_auth.git
cd banknote_auth
pip install pandas numpy matplotlib

🏗️ Project Structure

├── main.ipynb  # Jupyter Notebook with implementation
├── README.md   # Project Documentation

📊 Results

• Accuracy: ~90% on test data
• Max Depth: Configurable (default: 3)
• Visualization: Scatter plot of feature distribution

✨ Next Steps

• Implement pruning to prevent overfitting.
• Support entropy as an alternative impurity measure.
• Convert the code into a class-based implementation.

---

👨‍💻 Developed by: Kartik Agrawal