Stroke Risk Prediction Machine Learning Techniques

Abstract

Stroke is a critical medical condition and a leading cause of mortality and long-term disability across the globe. Accurate and early prediction of stroke risk can significantly enhance preventative care and reduce healthcare burdens. In this study, we propose a machine learning framework leveraging deep learning techniques to predict the likelihood of stroke using demographic, clinical, and lifestyle data. The model was developed using the TensorFlow/Keras framework and trained on a publicly available dataset. Key preprocessing steps included handling class imbalance with oversampling, feature normalization, and label encoding. A fully connected neural network architecture was implemented and optimized using techniques such as dropout and early stopping to prevent overfitting. The model achieved promising performance with high accuracy and AUC scores, demonstrating its capability to identify individuals at elevated stroke risk. Additionally, feature importance analysis using SHAP values provided interpretable insights into the most influential predictors. This research highlights the effectiveness of deep learning models in medical risk prediction and supports the integration of AI tools into clinical decision-making processes.