Deep Learning for Enhancing Diabetes Prediction

The WHO defines diabetes mellitus as a chronic illness characterized by elevated blood sugar levels and associated with far-reaching effects. Based on recent increases in mortality, diabetes is now the tenth most common cause of death across the globe. A wide range of unexpected results have been observed using machine learning (ML) classifiers and well-established methods for encoding categorical data to predict diabetes through imbalanced testing datasets. An artificial neural network has been used in earlier research to retrieve features without understanding the sequence information. The current study accurately predicts diabetes from patient data using bidirectional long/short-term memory (BiLSTM), a deep learning-based decision support system (DSS). The BiLSTM hybrid model is used to predict diabetes after the data set is balanced. In contrast to previous research, the trial results for this proposed model demonstrate improvement over existing methods, with a 90% rate of accuracy, precision, recall, and F1-score.