Heart Disease Prediction — Machine Learning Analysis

A machine learning project comparing five classification models for predicting heart disease risk using clinical patient data. Built as part of CMPT 353 at Simon Fraser University.

Overview

Heart disease is a leading cause of mortality worldwide. This project applies machine learning to a clinical dataset of 918 patients to predict heart disease risk, with the goal of identifying the most effective classification approach for early detection.

Dataset

• Source: Kaggle Heart Disease Dataset
• Size: 918 patients, 12 clinical attributes
• Target: Binary classification — presence or absence of heart disease

Features include: Age, Sex, ChestPainType, RestingBP, Cholesterol, FastingBS, RestingECG, MaxHR, ExerciseAngina, Oldpeak, ST_Slope

Methodology

Data Preprocessing

• Categorical encoding (binary and ordinal mappings)
• Feature normalization using MinMaxScaler
• Feature selection using SelectKBest (chi-squared test)
• Correlation analysis to identify and drop low-signal features
• Train/test split: 70% training, 30% testing

Key Features Identified

Based on correlation analysis and SelectKBest scoring, the strongest predictors were ST_Slope, ExerciseAngina, ChestPainType, and Oldpeak.

Models & Results

Model	Accuracy	Precision (Heart Disease)	Recall (Heart Disease)
Random Forest	88%	90%	90%
Gradient Boosting	87%	90%	88%
MLP Classifier	87%	89%	90%
Decision Tree	86%	91%	84%
K-Nearest Neighbours	83%	90%	79%

Best performer: Random Forest Classifier with 88% accuracy and balanced precision/recall across both classes.

Tech Stack

• Python
• pandas, numpy
• scikit-learn (GradientBoostingClassifier, RandomForestClassifier, MLPClassifier, DecisionTreeClassifier, KNeighborsClassifier)
• matplotlib, seaborn

Project Structure

heart-disease-ml/
├── heart-disease-prediction.ipynb   # Main Jupyter notebook
├── features-eng-and-ML-analyze.py   # ML model training and evaluation
├── visualize-and-normalize.py       # EDA and data preprocessing
├── heart.csv                        # Raw dataset
├── normalized_heart_data.csv        # Preprocessed dataset
├── Heart_Attack_Prediction_ML_Model_.pdf  # Full project report
└── *.png                            # Visualization outputs

Key Findings

• ST_Slope, ExerciseAngina, and ChestPainType were the strongest predictors of heart disease
• All models showed higher precision for positive (heart disease) cases than negative cases
• Random Forest and Gradient Boosting achieved the best overall balance of precision and recall
• Close training and validation scores across all models indicate well-tuned, generalizable models

Author

Gourav Bhardwaj — Simon Fraser University, CMPT 353