Heart Attack Risk Prediction

Project Structure

This repository contains a complete data science pipeline for predicting heart attack risk using machine learning for 4IZ565/4IZ566 – Programming for Data Science in Python. The project includes:

• Data preparation and feature engineering
• Exploratory data analysis and feature selection
• Model training, hyperparameter tuning, and evaluation

Files

• heart_attack_prediction_dataset.csv: Raw dataset with medical and demographic data
• heart_attack_feature_engineered.csv: Cleaned and enriched dataset after preprocessing
• data_preparation.ipynb: Data cleaning and feature engineering notebook
• exploratory_analysis_and_feature_selection.ipynb: EDA, visualizations, and feature selection notebook
• model_training_and_evaluation.ipynb: Model training, hyperparameter tuning, and evaluation notebook
• Visualizations:
  • correlation_heatmap.png: Heatmap of feature correlations
  • feature_histograms_continuous.png: Histograms of continuous features
  • feature_binary_counts.png: Count plots of binary features
  • rf_feature_importance.png: Feature importance from Random Forest
  • heart_attack_roc_curve_v2.png: ROC curve comparing model performance

---

Description of the Dataset

The dataset comes from https://www.kaggle.com/datasets/iamsouravbanerjee/heart-attack-prediction-dataset and contains medical and demographic data for 8,786 patients. The goal is to identify factors that influence heart attack risk and build predictive models.

Features include:

• Blood Pressure (Systolic/Diastolic)
• BMI and BMI Category
• Cholesterol and Triglycerides
• Pulse Pressure and Heart Rate
• Lifestyle (Exercise, Sedentary hours, Diet)
• Demographics (Age, Age Group, Sex, Continent, Income)

Target variable:

• Heart Attack Risk: 0 = No Risk, 1 = High Risk

---

Data Preprocessing

Performed in data_preparation.ipynb using pandas:

Key steps:

• Converted Blood Pressure string into numeric Systolic BP and Diastolic BP
• Created new features:

  • Cardio Risk Score: A composite health risk index calculated using a weighted formula based on five key factors known to influence cardiovascular risk:

    • Age
    • Sex (male = higher risk)
    • Systolic BP (systolic blood pressure)
    • Cholesterol (total cholesterol level)
    • (Originally included Smoking, but omitted if not available in dataset)

    The formula used:

    Cardio Risk Score = 
        0.03 × Age +
        1.0 × (1 if Male else 0) +
        0.02 × Systolic BP +
        0.01 × Cholesterol
    

  • Pulse Pressure = Systolic - Diastolic

  • BMI Category and Age Group

• Added geographic features: Patient count per continent
• Saved the output as heart_attack_feature_engineered.csv

---

Exploratory Data Analysis & Feature Selection

Performed in exploratory_analysis_and_feature_selection.ipynb

Key tasks:

• Separated features into categorical, binary, continuous, and numerical groups
• Visualizations:
  • correlation_heatmap.png: Correlation heatmap for numerical features
  • feature_histograms_continuous.png: Histograms for continuous variables
  • feature_binary_counts.png: Count plots for binary variables
• Feature selection:
  • RFE (Recursive Feature Elimination) using Logistic Regression
  • Feature importances using Random Forest (rf_feature_importance.png)

---

Modeling

Performed in model_training_and_evaluation.ipynb

Pipeline steps:

• Preprocessing:
  • Scaling numerical features
  • One-hot encoding categorical variables
• Classifiers:
  • Logistic Regression
  • Random Forest
  • Support Vector Classifier (SVC)
• Data split: 80/20 train/test
• Hyperparameter tuning with GridSearchCV
• Model selection based on test and cross-validation scores

---

Results and Evaluation

Accuracy (test set):

• Logistic Regression: 0.6418
• Random Forest: 0.6414
• SVC: 0.6418

ROC Curve:

• Saved as heart_attack_roc_curve_v2.png
• Displays AUC scores and performance comparison

Cross-Validation:

• 5-fold CV for each model
• Average accuracy printed for all

---

Notes and Design Decisions

• Feature selection aided by Random Forest and RFE
• Three different classifier types used for diversity
• Pipelines prevent data leakage and ensure reproducibility

---

Reproducibility Instructions

To run this project locally:

1. Clone this repository or download the files
2. Make sure heart_attack_prediction_dataset.csv is in your working directory
3. Open notebooks in this order:
   • data_preparation.ipynb
   • exploratory_analysis_and_feature_selection.ipynb
   • model_training_and_evaluation.ipynb
4. Run each notebook cell-by-cell to reproduce results

Dependencies

Make sure the following Python libraries are installed in your environment:

• pandas
• matplotlib
• seaborn
• scikit-learn

You can install them via:

pip install pandas matplotlib seaborn scikit-learn