SalahD8/stroke-prediction-model

Stroke Prediction using Machine Learning

Project Overview

This project develops a machine learning pipeline to predict the likelihood of strokes using the Kaggle Stroke Prediction Dataset. The goal is to accurately classify individuals into stroke or non-stroke categories while addressing challenges such as class imbalance.

Dataset

The dataset contains the following features:

• age: Age of the patient.
• avg_glucose_level: Average glucose level in the blood.
• bmi: Body Mass Index of the individual.
• gender, work_type, smoking_status: Categorical attributes describing patient demographics.
• stroke: Target variable indicating whether the patient had a stroke (1) or not (0).

Data Source

• Link: Kaggle Stroke Prediction Dataset

Workflow Pipeline

1. Data Preparation:

   • Missing values in bmi were imputed with the mean value.
   • Irrelevant columns like id were dropped.

2. Exploratory Data Analysis (EDA):

   • Visualized distributions of numerical features (age, avg_glucose_level, bmi).
   • Analyzed class distributions and correlations between features.

3. Feature Engineering:

   • Applied one-hot encoding for categorical variables.
   • Standardized numerical features for consistent scaling.

4. Model Development:

   • Implemented and evaluated multiple models:
     • Logistic Regression
     • Random Forest
     • XGBoost
     • Gradient Boosting
   • Addressed class imbalance using SMOTE to oversample the minority class.
   • Models were evaluated using metrics such as Accuracy, Precision, Recall, F1-Score, and ROC-AUC.

5. Software Engineering Best Practices:

   • Integrated unit tests to validate pipeline components.
   • Enhanced logging for debugging and workflow traceability.

6. Deployment:

   • [Optional] Model deployment details go here if you’ve deployed it.

Results

Model	Accuracy	Precision	Recall	F1-Score	ROC-AUC
Logistic Regression	0.75	0.60	0.40	0.48	0.82
Random Forest (SMOTE)	0.88	0.14	0.26	0.18	0.77
XGBoost (SMOTE)	0.89	0.13	0.22	0.17	0.78
Gradient Boosting (SMOTE)	0.80	0.14	0.60	0.23	0.79

Key Findings:

• Gradient Boosting with SMOTE performed best in detecting the minority class (stroke = 1), achieving the highest recall (0.60) and ROC-AUC (0.79).

Usage Instructions

Prerequisites

• Python 3.10.0
• Libraries: pandas, numpy, matplotlib, seaborn, scikit-learn, imbalanced-learn, xgboost

Steps to Run the Pipeline

1. Clone the repository: bash git clone https://github.com/your-repo/stroke-prediction.git cd stroke-prediction

2. Install dependencies: bash pip install -r requirements.txt

3. Run the pipeline: bash python pipeline.py

4. Run tests: bash pytest test_pipeline.py

Future Enhancements

• Hyperparameter tuning for better model performance.
• Deployment of the best model using Flask or FastAPI.
• Integration of additional datasets for improved generalization.

Contributors

• Salah Dabor