heart-attack-prediction

🫀 Heart Attack Risk Prediction

🌐 Live Demo: https://heart-attack-prediction-ilk1.onrender.com

ML Internship Project · IntrainTech, Bangalore · Aug–Nov 2023 Role: Machine Learning Engineer Intern

---

📌 What This Project Does

End-to-end heart attack risk prediction system — from raw clinical data to a live Flask web application and Power BI dashboard. Patients fill a form, the model predicts risk probability, and the app returns personalised lifestyle change recommendations.

---

🏗️ System Architecture

Patient fills web form (test.html)
            │
            ▼ POST /predict
┌──────────────────────────────┐
│         server.py            │
│                              │
│  1. Parse form inputs        │
│  2. Scale with StandardScaler│
│  3. model.predict_proba()    │
│  4. determine_lifestyle_     │
│     changes(prob, inputs)    │
│  5. Return JSON response     │
└──────────────┬───────────────┘
               │
               ▼
    result_template.html
    • Risk probability score
    • High / Low risk label
    • Personalised recommendations
      (smoking, BMI, exercise,
       diet, sleep, stress)

---

📊 Model Benchmark (10-Fold Cross-Validation)

Model	Accuracy
Random Forest ✅	69.17%
Light Gradient Boost	~67%
SVM	~65%
XGBoost	~64%
KNN	~63%
Logistic Regression	~62%
Decision Tree	~58%
Naive Bayes	~57%

Random Forest selected — best cross-validated accuracy across 10 folds. Evaluated using Accuracy, F1-Score, ROC-AUC, Precision, and Recall.

---

🔑 Key Engineering Decisions

Why SMOTE before training? Heart attack risk classes are imbalanced. SMOTE generates synthetic minority samples preserving feature distributions, preventing the model from always predicting the majority class.

Why StandardScaler? Features like Cholesterol (100–300), BMI (15–45), and Heart Rate (60–100) have very different ranges. Scaling ensures no single feature dominates distance-based calculations.

Why lifestyle recommendations? A risk score alone isn’t actionable. The recommendations engine maps specific input values (Smoking=1, BMI>25, Exercise<1.25h/week) to concrete changes — making the app clinically useful.

Why split Blood Pressure? The raw dataset stores BP as “120/80” string. Splitting into systolic and diastolic gives the model two meaningful numeric features instead of one useless string.

---

🗂️ Dataset

Heart Attack Risk Prediction Dataset — 8,763 patient records, 25 features:

Category	Features
Demographics	Age, Sex, Country, Continent, Hemisphere
Vitals	BP (Systolic/Diastolic split), Heart Rate, Cholesterol, BMI, Triglycerides
Lifestyle	Smoking, Alcohol, Exercise Hrs/Week, Diet, Sedentary Hrs, Stress Level, Sleep Hrs
Medical History	Diabetes, Family History, Previous Heart Problems, Obesity, Medication Use
Target	Heart Attack Risk (0 = Low, 1 = High)

---

🛠️ Tech Stack

---

📁 Project Structure

heart-attack-prediction/
├── .github/
│   └── workflows/
│       └── ci.yml                       # GitHub Actions CI
├── server.py                            # Flask app — trains model + serves predictions
├── 1.ipynb                              # Full EDA + 8-model benchmark notebook
├── heart_attack_prediction_dataset.csv  # Dataset (8,763 patient records)
├── Dashboard.pbix                       # Power BI dashboard
├── templates/
│   ├── test.html                        # Patient input form
│   └── result_template.html             # Risk result + lifestyle suggestions
├── requirements.txt
└── README.md

---

🚀 How to Run

# Clone the repo
git clone https://github.com/samuel-mekala/heart-attack-prediction.git
cd heart-attack-prediction

# Install dependencies
pip install -r requirements.txt

# Run the Flask app
# (model trains automatically on startup — ~10-15 seconds)
python server.py

# Open browser → http://localhost:5000

# To explore EDA and all 8 models:
jupyter notebook 1.ipynb

---

🔮 Future Work

• Deploy on Render for public access
• Add SHAP explainability — show which features drive each prediction
• REST API endpoint for hospital management system integration
• Patient history tracking with database backend
• Retrain pipeline with new patient data

---

IntrainTech Internship · Bangalore · Aug–Nov 2023