🧠 Machine Learning-Based Stroke Prediction Model for Early Detection and Prevention

📊 Research Overview

This research project implements a stroke risk prediction system using an Artificial Neural Network (ANN) model integrated with a user-friendly Streamlit web interface. The system analyzes personal health metrics to assess an individual's risk of experiencing a stroke, providing a valuable tool for preventive healthcare.

B.Sc. Research Project in Electrical and Electronic Engineering - Sri Lanka Institute of Information Technology (SLIIT)

---

🔍 Research Highlights

✅ Dataset Characteristics

• 8,600 health records collected from hospital data
• 2,500 stroke cases identified for analysis
• Comprehensive patient health profiles including medical history and lifestyle factors

✅ Advanced Neural Network Architecture

• Implemented a deep neural network with multiple hidden layers (1024→512→256→128→64→32→16→1)
• Incorporated dropout layers (0.2) to prevent overfitting
• Applied L2 regularization for model stability
• Utilized Adam optimizer with fine-tuned learning rate

✅ Data Processing & Model Training

• Preprocessed healthcare dataset (categorical encoding, outlier handling)
• Resolved data imbalance using SMOTE (Synthetic Minority Over-sampling Technique)
• Implemented feature scaling using StandardScaler
• Achieved 81% accuracy in stroke prediction

✅ Model Performance Metrics

• Precision: ~81% - High reliability in positive predictions
• Recall: ~82% - Strong capability to identify actual stroke cases
• F1-Score: ~81-82% - Balanced performance between precision and recall

✅ Interactive Web Application

• Developed a responsive Streamlit interface for user interaction
• Created an intuitive health information form with appropriate input constraints
• Implemented real-time risk assessment with visual feedback
• Displayed prediction results with probability scores

---

💻 Technologies Used

• Python - Core programming language
• TensorFlow/Keras - Neural network implementation
• Pandas - Data manipulation and preprocessing
• NumPy - Numerical operations
• Scikit-learn - Machine learning utilities (SMOTE, StandardScaler)
• Matplotlib/Seaborn - Data visualization
• Streamlit - Web application framework
• Pickle - Model serialization

---

📋 Key Health Indicators Analyzed

The ANN model processes the following user health metrics to predict stroke risk:

Feature	Type	Description	Significance
Gender	Categorical	Male/Female	Different risk profiles by gender
Age	Numerical	Age in years	Increasing risk with age
Hypertension	Binary	0=No, 1=Yes	Major stroke risk factor
Heart Disease	Binary	0=No, 1=Yes	Comorbidity affecting stroke risk
Avg. Glucose Level	Numerical	Blood glucose level (mg/dL)	Diabetes-related risk indicator
BMI	Numerical	Body Mass Index	Weight-related risk factor
Smoking Status	Categorical	Never smoked/Formerly smoked/Smokes	Lifestyle risk factor

---

🖼️ Application Interface

The interactive Streamlit application provides an intuitive interface for users to input their health data and receive stroke risk predictions.

---

🚀 Installation & Usage

1. Clone the repository

   git clone https://github.com/yourusername/stroke-prediction-system.git
   cd stroke-prediction-system

2. Install dependencies

   pip install -r requirements.txt

3. Run the Streamlit application

   streamlit run app.py

4. Access the web interface Open your browser and navigate to http://localhost:8501

---

📈 Model Performance

Metric	Value
Accuracy	81%
Precision	81%
Recall	82%
F1-Score	81-82%

---

📚 Implementation Details

The system workflow consists of three main components:

1. Data Preprocessing Pipeline

   • Categorical feature encoding
   • Missing value imputation
   • Feature scaling
   • Class imbalance handling via SMOTE

2. Neural Network Model

   model = Sequential([
       Dense(1024, activation='relu', kernel_regularizer=l2(0.01)),
       Dropout(0.2),
       Dense(512, activation='relu', kernel_regularizer=l2(0.01)),
       Dropout(0.2),
       # Additional layers...
       Dense(1, activation='sigmoid')
   ])

3. Web Application Integration

   • User data collection
   • Real-time prediction
   • Result visualization

---

🔮 Future Research Directions

• Model Refinement: Further optimization of neural network architecture
• Hyperparameter Tuning: Systematic exploration of optimal parameters
• Clinical Deployment: Testing in real-world healthcare settings
• Feature Expansion: Incorporate additional health parameters for improved prediction
• Explainability: Implement feature importance visualization for clinical interpretation
• Personalized Recommendations: Add tailored health advice based on risk factors
• Mobile Application: Develop companion app for wider accessibility
• Longitudinal Analysis: Implement user accounts for tracking health metrics over time

---

🌟 Research Impact

This research contributes to AI-driven healthcare solutions by:

• Enhancing Early Detection: Identifying high-risk individuals before symptom onset
• Supporting Preventive Care: Enabling targeted interventions for at-risk populations
• Reducing Healthcare Burden: Potential reduction in emergency care costs through prevention
• Democratizing Healthcare: Providing accessible stroke risk assessment tools