Author(s): Dr. G Kalpana, PVS Lakshmi, M Likhitha Sree and Asra Mohammad

Abstract: Stroke is a serious medical condition that develops when there is a disruption in the flow of blood to the brain, leading to neurological damage. It is a major global danger with detrimental effects on both health and the economy. In order to address this, scientists are creating automated algorithms for predicting strokes, which might potentially prevent fatalities and enable early intervention. With the aging of the population comes an increased risk of stroke, therefore accurate and reliable prediction methods become more and more important. In order to better understand the decision-making process involved in balancing the dataset from a ratio of 19:1 for No Stroke: Stroke to an equal ratio via SMOTE Analysis, the study set out to accomplish three goals: develop an accurate machine learning model for predicting stroke disease; address the serious class imbalance issue that arises from the a stroke patients' school being significantly smaller compared to the healthy class; and find the import feature using Mutual Information rating, Chi-Square Rating, and ANOVA test. Using an Android app, provide an end-to-end intelligent healthcare system. The efficacy of the suggested machine learning method was investigated using metrics pertaining to prediction accuracy and generalization capacity in a comparative analysis with six popular classifiers. We also looked at two other explainable method types in this study: SHAP and LIME, to provide insight into these black-box machine learning models. Especially in the medical field, well-proven and trustworthy methods for elucidating model decision-making include SHAP (Shapley, for Additive Explanations) or LIME (Local Comprehensible Model-agnostic Explanations). According to the experiment results, more complex models performed better than simpler ones; the best model had an accuracy of almost 91%, while the other models had an accuracy of 83-91%). By standardizing complex models and providing knowledge about their decision-making, the suggested framework—which incorporates both global and local understandable methodologies—can improve stroke treatment and care.

DOI: 10.33545/27076571.2024.v5.i2b.105

Pages: 124-128 | Views: 879 | Downloads: 355

Download Full Article: Click Here