GAN

REPO LINK HERE Strokes represent a major global health concern, being one of the leading causes of disability worldwide. As medical science advances, the ability to predict and prevent stroke occurrences through early detection has become increasingly crucial. Strokes resulting from an abrupt interruption of blood flow to the brain, can lead to permanent brain damage or death. With the incidence of stroke doubling in the last three decades, and an estimated one stroke occurring every three seconds around the globe, it is clear that this is a condition that affects individuals across all age groups, not just the elderly. Recognizing the key indicators of high stroke risk in patient health records can significantly aid in mitigating this trend. Through interventions such as lifestyle modifications and targeted medical treatments, healthcare providers can dedicate resources more effectively to reduce the occurrence of strokes and manage their aftermath, ultimately improving patient outcomes and reducing the burden on health systems. The focus of this project is to meticulously analyze various algorithms to identify the most effective approach for predicting stroke occurrences. In the quest to address this pressing medical challenge, the report will explore different machine learning models, utilizing patient health records from Kaggle [1] as a dataset. The objective is to pinpoint an algorithm that excels in accuracy, sensitivity, and specificity, thereby enabling healthcare providers to intervene proactively and allocate medical resources efficiently to at-risk individuals. By leveraging data-driven insights, it is hoped that the algorithm can contribute to the global efforts in reducing stroke incidences and the associated healthcare burdens. Methods explored: EDA Oversampling techniques with (SMOTE, ADSYN) Data Augmentation with cGAN ML prediction (Logistics regression, random forest, gradient boosting, SVM) Deep learning (hyperparmeter tuning with neural networks)

Nov 20, 2023