Project 2: Decision Tree Classifier for Real-World Datasets

Overview

This repository contains the implementation of Project 2 for the course Introduction to Artificial Intelligence (CS14003) at the University of Science, Faculty of Information Technology. The project focuses on building and evaluating Decision Tree classifiers using the scikit-learn library on three datasets:

• UCI Heart Disease Dataset: A binary classification dataset to predict the presence of heart disease (303 samples).
• Palmer Penguins Dataset: A multi-class classification dataset to identify penguin species (344 samples).
• Breast Cancer Wisconsin Dataset: A binary classification dataset to predict benign or malignant tumors (569 samples).

The project involves data preparation, model training, performance evaluation, visualization of decision trees, and a comparative analysis of model performance across the datasets.

Table of Contents

1. Project Structure
2. Requirements
3. Setup Instructions
   • Using pip
4. How to Run
5. Datasets
6. Results
7. Team Members
8. License

Project Structure

project2_decision_tree/
│
├── notebooks/
│   ├── heart_disease.ipynb       # Analysis for UCI Heart Disease dataset
│   ├── palmer_penguins.ipynb    # Analysis for Palmer Penguins dataset
│   ├── breast_cancer.ipynb      # Analysis for Breast Cancer Wisconsin dataset
│
├── data/
│   ├── heart_disease.csv        # UCI Heart Disease dataset (optional)
│   ├── penguins.csv             # Palmer Penguins dataset (optional)
│   ├── breast_cancer.csv        # Breast Cancer Wisconsin dataset (optional)
│
├── report/
│   ├── project_report.pdf       # Final report with analysis and insights
│
├── README.md                    # This file
└── requirements.txt             # Dependencies for pip

Requirements

The project requires Python 3.8+ and the following libraries:

• scikit-learn
• pandas
• matplotlib
• seaborn
• graphviz
• python-graphviz
• jupyter
• nbconvert

Setup Instructions

Using pip

1. Install Python:

   • Download Python 3.8+ from python.org.

   • Verify:

     python --version

2. Create a virtual environment (recommended):

   python -m venv decision_tree_env
   source decision_tree_env/bin/activate  # On Windows: decision_tree_env\Scripts\activate

3. Install dependencies:

   pip install -r requirements.txt

   Or manually:

   pip install scikit-learn pandas matplotlib seaborn jupyter graphviz nbconvert

4. Install Graphviz:

   • Download from graphviz.org.

   • Add Graphviz to PATH (e.g., C:\Program Files\Graphviz\bin on Windows).

   • Verify:

     dot -V

5. Verify libraries:

   import sklearn, pandas, matplotlib, seaborn, graphviz
   print("Libraries installed successfully!")

How to Run

1. Clone the repository:

   git clone https://github.com/Low-rain-falls/CSC14003_Decision_Tree.git
   cd CSC14003_Decision_Tree

2. Activate the environment:

   • pip:

     python3 -m venv venv
     source venv/bin/activate

3. Launch Jupyter Notebook:

   jupyter notebook

4. Run notebooks:

   • Open notebooks/heart_disease.ipynb, palmer_penguins.ipynb, or breast_cancer.ipynb in the Jupyter interface.
   • Execute all cells to perform the analysis.

5. Export to PDF (optional):

   jupyter nbconvert --to pdf notebooks/[notebook_name].ipynb

Datasets

1. UCI Heart Disease:

   • Source: UCI Machine Learning Repository
   • Description: 303 samples, 13 features, binary classification (0: no heart disease, 1: heart disease).
   • Access: Loaded via URL or data/heart_disease.csv.

2. Palmer Penguins:

   • Source: seaborn or Palmer Penguins
   • Description: 344 samples, 6 features, 3 classes (Adelie, Chinstrap, Gentoo).
   • Access: Loaded via seaborn.load_dataset('penguins') or data/penguins.csv.

3. Breast Cancer Wisconsin:

   • Source: UCI Machine Learning Repository
   • Description: 569 samples, 30 features, binary classification (benign or malignant).
   • Access: Loaded via sklearn.datasets.load_breast_cancer() or data/breast_cancer.csv.

Results

• Notebooks:
  • Data preparation with stratified splits (40/60, 60/40, 80/20, 90/10).
  • Decision tree training and visualization using Graphviz.
  • Evaluation with classification reports and confusion matrices.
  • Analysis of max_depth impact on accuracy (80/20 split).
• Report (report/project_report.pdf):
  • Visualizations and statistical results.
  • Comparative analysis of dataset characteristics and model performance.
  • Insights on decision tree behavior across datasets.

License

This project is for educational purposes only. Datasets are used under their respective public licenses.