Early Diabetes Prediction Using CNN-LSTM and CNN-Bi-LSTM Models Optimized with Adam on the PIMA Indian Dataset

Early detection of the condition is critical to proper diabetes management and reducing risk for complications. A new concept for early prediction of diabetes is proposed in this study where the LSTM and CNN models are combined with Bi-LSTM. The target is to use the Adam optimizer for these models on top of improving this model, and testing it onto the PIMA Indian Diabetes Dataset. The CNNLSTM model is particularly well-suited for time dependent data as it has been designed to extract the spatial patterns from an input and in turn record memory of a problem across time steps. Meanwhile, with the data from both past and future, the CNN-Bi-LSTM model is capable of better prediction. From our experimental results, we can see that these hybrid models achieve superior performance over traditional machine learning algorithms and hence could serve as a powerful resource for predicting diabetes at an early stage. Thus the application of this approach is a new challenge and seems to have potential for improving early diagnosis in order to implement timely interventions in diabetes care.