Deep Learning Approaches to Heart Stroke Prediction: Model Evaluation and Insights

In recent times, it is critical for people who have suffered heart stroke to have a prognosis that is both accurate and timely. This will allow doctors to begin treatment at an earlier stage, which will significantly reduce the number of fatalities. This study examines the performance of various deep learning models with the intention of achieving a more accurate prediction of strokes. The study discusses various deep learning architectures, including CNN-1D, LSTM, CNN, Deep Belief Networks, CNN, and CNN-LSTM mixed models, in the context of heart stroke prediction. The models were trained and evaluated using performance metrics in terms of accuracy, confusion matrix evaluations, and ROC-AUC scores. These metrics were utilized for the purpose of training and evaluation. To obtain this information, a comprehensive dataset that had information on medical, behavioural, and demographic factors was collected and used for experimentation. Comparatively, the DBN model proved to be the most successful, with 90% accuracy and an area under the curve (AUC) value of 0.89. In addition to this, it had the lowest percentages of making classifications that were inaccurate. Both DNN and CNN-1D demonstrated that they are competitive in terms of performance, as evidenced by their accuracy of 82% and their AUC score of 0.82. On the other hand, LSTM and CNN-LSTM demonstrated a degree of performance that was moderate, with AUC values of 0.86 and 0.79, respectively. This is because the findings suggest that deep learning architectures proved effective for stroke prediction or applications on the similar datasets. Future scope of the presented work will be directed at improving the overall performance of the architectures through the implementation of complex ensemble techniques, the selection of features, and the resolution of the problem of class imbalance.