🩺 Heart Failure Prediction using Machine Learning

Educational project: This repository is for machine-learning practice and exploratory analysis. It is not a clinical diagnostic system and should not be used for medical decisions.

📌 Context & Motivation

Cardiovascular diseases (CVDs) are the leading cause of death globally. Early detection of heart conditions can significantly improve treatment outcomes and save lives. This project leverages machine learning to enable efficient and reliable heart disease prediction, aiding healthcare professionals in early diagnosis.

---

❓ Problem Statement

Heart disease often remains undiagnosed until it's too late, particularly in individuals with risk factors like high blood pressure, diabetes, or high cholesterol. The aim is to create a predictive model that can analyze patient health data and accurately detect the likelihood of heart disease before it becomes critical.

---

🎯 Project Aim

To develop a machine learning model that uses patient medical attributes to predict heart disease. This can assist medical professionals in identifying at-risk patients and providing early interventions.

---

🧾 Dataset Overview

Feature	Description
Age	Age of the patient (in years)
Sex	Biological sex (M: Male, F: Female)
ChestPainType	Type of chest pain (TA, ATA, NAP, ASY)
RestingBP	Resting blood pressure (mm Hg)
Cholesterol	Serum cholesterol (mg/dl)
FastingBS	Fasting blood sugar (1 if > 120 mg/dl, else 0)
RestingECG	Resting electrocardiogram results (Normal, ST, LVH)
MaxHR	Maximum heart rate achieved
ExerciseAngina	Exercise-induced angina (Y/N)
Oldpeak	ST depression induced by exercise
ST_Slope	Slope of peak exercise ST segment (Up, Flat, Down)
HeartDisease	Target variable (1: disease, 0: normal)

---

📒 Project Workflow

1. Data Loading and Initial Exploration

• Import libraries
• Load dataset into heart_failure_df
• Basic exploration and structure analysis

2. Preliminary Visualization & Feature Selection

• Visual distribution
• Correlation and medical relevance review

3. Data Cleaning

• Handle missing values & duplicates
• Detect & treat outliers
• Binning & smoothing

4. Feature Engineering

• Post-cleaning selection: Correlation matrix, Chi-square test, domain logic
• Normalization vs. Standardization using QQ Plot

5. Feature Encoding

• One-Hot Encoding for categorical and binned features

6. Data Splitting & Scaling

• Train-test split
• Apply normalization

---

📉 Dimensionality Reduction & Clustering

7. PCA (Principal Component Analysis)

• 2D PCA for visualization
• PCA retaining 95% variance for modeling

8. Clustering (Unsupervised Learning)

K-Means

• Elbow Method
• Dunn Index, Silhouette Score, Davies-Bouldin Index
• Cluster visualization

Hierarchical Clustering

• Dendrogram analysis
• Agglomerative clustering with optimal k=2

---

🤖 Classification Models

9. Model Training & Evaluation (With & Without PCA)

• Logistic Regression
• Random Forest Classifier
• Support Vector Machine (SVC)

Each model is evaluated using:

• Accuracy, Precision, Recall, F1-Score
• Decision Boundary visualization

10. Model Comparison

• Side-by-side performance analysis

---

🔧 Hyperparameter Tuning

11. Optimization Process

• K-Fold Cross-Validation
• GridSearchCV for hyperparameters
• Training & re-evaluation of:
  • Logistic Regression
  • Random Forest
  • SVC

12. Final Comparison

• Default vs. Tuned Models (with & without PCA)
• Final Decision Boundaries
• Feature Importance plots

---

📊 Results Summary

• Strong performance improvements from feature engineering and PCA
• Meaningful insights from clustering
• Optimized models outperform baselines
• Important medical features identified through feature importance analysis

---

🛠️ Technologies Used

• Python
• Pandas, NumPy
• Scikit-learn
• Matplotlib, Seaborn
• Jupyter Notebooks

---

📊 Model Comparison

The notebook evaluates the models on a held-out test set of 184 samples. The strongest reported configuration by accuracy and F1-score is tuned Random Forest with PCA.

Configuration	Accuracy	Precision	Recall	F1-score
Logistic Regression	89.67%	91.92%	89.22%	90.55%
Random Forest	84.24%	87.63%	83.33%	85.43%
Linear SVC	89.67%	91.09%	90.20%	90.64%
Tuned Linear SVC	90.22%	92.00%	90.20%	91.09%
Tuned Random Forest + PCA	90.22%	89.62%	93.14%	91.35%

Selected model: Tuned Random Forest with PCA, because it reports the highest recall and F1-score among the evaluated configurations while matching the best accuracy.

These results are from the notebook’s documented train/test split and should be interpreted as educational model evaluation, not clinical validation.

📁 Project Structure

heart-failure-prediction/ │ ├── data/ # Dataset files ├── notebooks/ # Jupyter Notebooks for EDA and modeling ├── models/ # Saved models (if applicable) ├── outputs/ # Plots and reports ├── README.md # Project documentation └── requirements.txt # Python dependencies

---

✅ Conclusion

This project demonstrates an end-to-end educational machine-learning workflow: preprocessing, feature engineering, dimensionality reduction, clustering, classification, and hyperparameter tuning. The reported metrics describe performance on this dataset and split only; the project is not a clinical diagnostic system and does not establish medical effectiveness.