Stroke Prediction Using Machine Learning

This repository contains a machine learning project aimed at predicting the risk of stroke in individuals using health-related data. The analysis includes data cleaning, exploratory data analysis (EDA), model development, and performance evaluation.

Project Objectives

• Perform exploratory data analysis (EDA) to identify key factors associated with stroke.
• Develop a baseline classification model to benchmark performance.
• Build and tune machine learning models to predict stroke risk.
• Evaluate models using metrics such as accuracy, recall, precision, and F1-score.
• Provide business recommendations based on insights from the analysis.

Dataset

The dataset used is publicly available and contains medical and demographic features including:

• Age
• Hypertension
• Heart Disease
• Work Type
• Smoking Status
• BMI
• Glucose Level
• Stroke Occurrence (target)

Exploratory Data Analysis (EDA)

EDA was conducted to uncover patterns, correlations, and outliers in the data:

• Distribution of stroke across age, gender, and health conditions
• Correlation between features
• Class imbalance analysis
• Visualizations include bar plots, histograms, and correlation heatmaps

Modeling and Evaluation

Several models were evaluated including:

• Logistic Regression (with and without SMOTE)
• Random Forest Classifier
• Hyperparameter tuning using GridSearchCV
• Performance metrics: Accuracy, Recall, Precision, F1 Score, ROC-AUC

Best Performing Model

• Logistic Regression (SMOTE + Tuning)
• Recall (stroke): 0.84
• Precision (stroke): 0.12
• F1-score (stroke): 0.22

Note

Recall was prioritized over precision to reduce false negatives in stroke detection.

Key Results

• Significant class imbalance addressed using SMOTE.
• High recall in stroke prediction shows potential for early risk flagging.
• False positives were increased, indicating a tradeoff for higher sensitivity.

Business Recommendations

• Use the model to flag at-risk individuals for further medical screening.
• Collect additional features (e.g., cholesterol, physical activity) to enhance model accuracy.
• Monitor and regularly update the model with new data to improve performance.

Next Steps

• Feature engineering (interaction terms, polynomial features).
• Implement cost-sensitive learning.
• Integrate SHAP for model interpretability.
• Consider deploying the model via a simple web app.