Advanced stroke prediction using machine learning and deep learning techniques

Abstract

Stroke remains a leading cause of disability and mortality, necessitating reliable predictive models for early diagnosis. This thesis enhances stroke prediction using a comprehensive Kaggle dataset of 43,400 clinical records, employing both machine learning and deep learning techniques. The methodology includes extensive data preprocessing, handling missing values, data imputation, encoding categorical variables, normalization, and addressing class imbalances, followed by the implementation and evaluation of algorithms such as Random Forest, Decision Tree, Support Vector Machine, Logistic Regression, K-Nearest Neighbors, and Deep Neural Networks (DNNs). Metrics like accuracy, precision, recall, F1-score, AUC, and specificity were used for evaluation. Ensemble methods like Gradient Boosting (G-Boost) and Extreme Gradient Boosting (XGB) outperformed others, with G-Boost achieving an accuracy of 99% using a 70-20-10 split ratio. DNN also showed promise in handling complex data. The contributions highlight the potential of advanced techniques in stroke prediction, achieving superior performance and providing insights into health parameters influencing stroke risk. This study contributes to medical informatics by setting benchmarks for stroke prediction and demonstrating AI's applicability in healthcare. Implementing these models in clinical settings can enhance early detection, enable personalized interventions, and improve patient outcomes, reducing the burden of stroke.