Publication

Abstract : Diabetes mellitus (DM) is a metabolic illness affecting millions of people over the globe and it results in severe diseases such as diabetic nephropathy, cardiac stroke, etc. So, the earlier detection of DM becomes essential to avoid the progression of disease to the advanced stage and damage other organs. The recently developed deep learning (DL) models find useful for accurate and prompt disease diagnosis processes. This paper designs an optimal bidirectional long short term memory based predictive analysis and severity estimation (OBLSTM-PASE) model for DM. The major aim of the OBLSTM-PASE technique is to determine the presence of DM and identify the severity level properly. Besides, the OBLSTM-PASE technique involves two major stages namely OBLSTM based DM prediction and density-based spatial clustering of applications with noise (DBSCAN) based severity level estimation. In addition, the optimal hyperparameter selection of the BLSTM model takes place using the salp swarm algorithm (SSA) and thereby improves the overall prediction performance. To validate the better outcomes of the OBLSTM-PASE technique, a set of experiments were carried out on PIMA Indians Diabetes dataset. A comprehensive comparative analysis pointed out the improved performance of the OBLSTM-PASE technique over the other approaches.