Autonomous prediction of Type 2 Diabetes with high impact of glucose level

Abstract

Cognitive Impairment (CI) is a syndrome in which a person's mental abilities are noticeably weakened when compared to others of the same age. Memory loss, loss of attention, language challenges, and difficulty making decisions and planning are some of the mental and thinking problems that people with cognitive impairment. One of the disorders caused by CI is Alzheimer's disease (AD). Previous research studies show that there is an association between diabetes and Alzheimer's disease. It is critical to identify the AD at beginning stage and there is no medication available. Severity of diabetes may lead to Alzheimer. Diabetes is said to be the most prevalent chronic malady and type II diabetes is high penetrance in human around the world. This will cause other complications such as heart failure, kidney failure, blindness, etc. The high and low blood glucose level of the diabetes patients may lead to dysfunction and weakening the organs such as nerves, blood vessels and eyes. Early prediction of diabetes can help us to avoid these complications. In this research, the diabetes dataset is preprocessed with the series of four stages such as normalization, weight adjustment method to treat the outlier, dimensionality reduction using t-distributed Stochastic Neighbor Embedding (tSNE) and balancing the imbalanced dataset using Synthetic Minority Over-sampling Technique (SMOT) which will avoid the overfitting problem. Among the features related to diabetes, blood glucose level is the most relevant feature used to identify type II diabetes. Likewise, the features related to diabetes prediction is selected using feature importance method. Datasets with most relevant features are used for classification using proposed convolution Graph Long Short-Term Memory (CGLSTM). The weight of this deep neural network is further optimized with AdaGrad optimizer to increase the accuracy of prediction. The proposed work is evaluated on Pima Indians Diabetes Database(PIDD). The experimented results are compared with already available approaches to demonstrate proposed system efficiency. The outcomes of the proposed Type II DM are promising compared to conventional classification systems.