A Hybrid Artificial Intelligence Approach for Early Stroke Prediction

This research aims to explore the application of Machine Learning (ML) models, specifically ensemble methods, in conjunction with Genetic Algorithms (GA) for the early prediction of strokes using real-world data from Kaggle. The primary focus is on enhancing the performance of the stacking ensemble method.Stacking leverages the strengths of various ML models while mitigating their weaknesses, creating a meta-model that aggregates predictions from base models. We conducted extensive testing involving multiple meta-models and compared their performance. The Multi-Layer Perceptron (MLP) meta-model, with tuned parameters, outperformed others. It was trained on selected dataset features and data from base models selected via GA. Subsequently, we assessed its performance using an independent test set. We employed various metrics for evaluation, including accuracy, F1 score, recall, precision, AUC-ROC, and Matthews Correlation Coefficient (MCC). Our assessment also encompassed measuring time complexity by calculating execution time and prediction time. The results underscore the superior performance of this meta-model across all metrics, including accuracy (99.69%), F1-score (99.68%), recall (99.78%), precision (99.57%), AUC-ROC (99.69%), MCC (99.38%), execution time (2.20 minutes), and prediction time (0.0020 seconds).