🧠 Stroke Risk Prediction

📌 Project Overview

Stroke is a critical health condition and a leading cause of death and long-term disability worldwide. This project focuses on predicting the risk of stroke using machine learning models based on medical and demographic features.

Two models were implemented and evaluated:

• Decision Tree Classifier
• Random Forest Classifier (🏆 Best Performing)

---

📂 Project Resources

🔹 Dataset (CSV File): Download healthcare-dataset-stroke-data.csv
🔹 Project Code (.ipynb): View Jupyter Notebook
🔹 Project Report (DOCX File): Download Report
🔹 Presentation (PPTX File): Download PPT

---

🧾 Dataset Overview

• Total Records: 5,110
• Target Variable: stroke (1 = Stroke occurred, 0 = No stroke)
• Features:
  • Demographic: Age, Gender, Marital Status, Residence Type
  • Clinical: Hypertension, Heart Disease, Glucose Level, BMI
  • Lifestyle: Smoking Status, Work Type

📝 Class Imbalance Notice: Only ~5% of the data points represent stroke cases, making class balancing necessary.

---

🛠 Data Preprocessing

✔️ Cleaning & Transformation:

• Dropped non-informative id column
• Handled missing values in bmi using median imputation
• Encoded categorical variables using LabelEncoder

✔️ Outlier Removal:

• Used IQR method to eliminate outliers in bmi and avg_glucose_level

✔️ Feature Scaling:

• Applied StandardScaler to numerical columns

✔️ Train-Test Split:

• 80% Training, 20% Testing using Stratified Sampling

✔️ Imbalance Handling:

• Used SMOTE (Synthetic Minority Over-sampling Technique) on training set

---

📊 Exploratory Data Analysis

• Visualizations: Histograms, KDE plots, bar charts, and heatmaps
• Findings:
  • Higher age and glucose levels correlate positively with stroke risk
  • Hypertension and heart disease increase likelihood of stroke
  • Gender and residence type have minimal impact

---

🤖 Models Used

1️⃣ Decision Tree Classifier

✔️ Transparent and interpretable
❌ Slightly lower accuracy on imbalanced data

2️⃣ Random Forest Classifier (🏆 Best Model)

✅ High accuracy and better generalization
✅ Handles imbalance effectively with SMOTE
✅ Offers feature importance insights

---

📈 Model Performance

Model	Accuracy	Precision	F1 Score
Decision Tree	89.5%	68.2%	74.4%
Random Forest	93.7%	75.9%	81.1%

📌 Random Forest outperformed Decision Tree in all metrics.

---

🔍 Feature Importance

The top predictors identified by the Random Forest model:

• Age 🥇
• Average Glucose Level
• BMI
• Hypertension
• Heart Disease
• Smoking Status

---

🚀 Future Scope

🔹 Incorporate more clinical features (blood pressure, cholesterol, medication history)
🔹 Apply deep learning methods (e.g., Neural Networks, LSTM)
🔹 Develop a real-time decision support system for healthcare providers

---

🏗 Project Highlights

This project showcases the complete data science workflow for stroke risk prediction, including:
✔️ Data Preprocessing & Cleaning
✔️ Exploratory Data Analysis and Feature Engineering
✔️ Model Development and Evaluation (Decision Tree, Random Forest)
✔️ Report, Visualization, and Documentation

---