Deep Learning-Based Continuous Glucose Monitoring with Diabetic Prediction Using Deep Spectral Recurrent Neural Network

Abstract

It is estimated that approximately 50% of the world's population has diabetes mellitus. Diabetic diseases are caused by either a lack of insulin produced by the pancreas or a lack of insulin used efficiently by the body. Every year, a lot of money is spent on treating a person with diabetes on. An individual with diabetes has either insufficient insulin produced by the pancreas or ineffective utilisation of insulin by the body, resulting in chronic disease. Every year, a lot of money is spent on treating a person with diabetes. Therefore, prediction is the most important issue and the most accurate and reliable application method. It also needs to be more precise in determining patients’ insulin levels. To overcome this problem, this study proposes an approach using the deep spectral recurrent neural network (DSRNN) algorithm. It is one of the artificial intelligence systems, especially the deep spectral recurrent neural network (DSRNN), used to predict insulin levels in diabetic patients. Deep spectral recurrent neural networks (DSRNN) were selected to develop models for predicting diabetes. Initially, using the diabetic dataset for analysis, the expected result is based on training and testing processing. Then, preprocessing is used to reduce the irrelevant data. Preprocessed data will enter the next step of feature extraction using the multiscalar feature selection (MSFS) algorithm. In this method of analysis, the data is based on maximum weights. And they selected the features' threshold values using social spider optimisation (SSO) analysis of the importance of the features. Finally, enter the classification using DSRNN for a diabetic prediction based on the insulin level. Diabetes technology, such as continuous glucose monitoring (CGM), provides a wealth of data that enables measurement. Depending on the technology used, the sampling frequency varies in minutes. This model is a good predictor, and the probability model for diabetes is tested with accuracy on experimental data. Higher accuracy can be achieved if models are trained on future data.