Enhancing Multi-Disease Prediction using Deep Learning and Convergence based Optimization

Prediction of diseases at an early stage with the help of symptoms is a critical part of intelligent healthcare. This paper presents a deep learning framework to predict diseases in multiple classes with the help of a large-scale set of symptoms-disease data including 246,945 patient records, 377 symptoms, and 773 diseases. Following processing and cleaning of data, 658 disease classes were then maintained to be used in development of a model. Convolutional Neural Networks (CNN), Artificial Neural Networks (ANN), and Graph Neural Networks (GNN) have been compared in the same conditions of the experiment. Models were not trained with fixed training epochs but trained with a convergence-based optimization strategy with early stopping, which automatically stops when the validation performance begins to stagnate. Experimental findings show that ANN model attained the highest level of validation with a 0.8599 accuracy which is supported by the confusion matrix analysis which indicates that the ANN model had low misclassification rates. The results underscore the significance of preprocessing and convergence-directed training in ensuring dependable large-scale prediction of diseases systems.