Application of Soft Computing for Prediction of Peripheral Neuropathy in T2DM-A Real-Time Data Analysis

Type 2 Diabetes mellitus (T2DM) patients are experiencing diabetic foot problems that heavily burden healthcare systems around the globe. Although there have been challenges with diagnosing and treating these problems using traditional approaches, the advent of machine learning technology signals the beginning of a new age in diabetic foot care, with the promise of improved precision and customized treatment plans. Machine learning is a beneficial tool for extracting essential insights from large, complicated datasets to improve the accuracy of diabetic foot diagnosis and therapeutic planning. This research aims to employ artificial intelligence to build a decision support system that will use clinical and demographic variables to predict the likelihood that individuals with mild, moderate, or severe peripheral neuropathy may develop diabetic foot syndrome in T2DM. Real-time processing of clinical information is made possible by a customized stacked ensemble model, which offers immediate peripheral neuropathy risk prediction with low computing latency. The system's capacity to transform raw patients' data into valuable insights in milliseconds supports quick clinical decision-making. Additionally, comparison and testing have been conducted on four deep learning algorithms: ResMLP, LSTM, DNN, and 1D-CNN. The predictions produced by the classifiers have been interpreted using three explainers: LIME, Eli5, and SHAP. Using the mutual information feature selection technique, the final stack reached a maximum accuracy of 99%. The three most significant markers that helped predict the onset of diabetic foot syndrome were area under pressure, vpt right, and vpt left; the encouraging findings point to the possibility of predicting diabetic foot condition with a decision system.