Bio-Inspired PSO for Improving Neural Based Diabetes Prediction System

Abstract

A high level of glucose in the blood over a long period creates diabetes disease. Undiagnosed diabetes may trigger other complications such as cardiovascular disease, nerve damage, renal failure, and so on. There are many factors age, blood pressure, food habits, lifestyle changes are some of the reasons for diabetes. With increasing cases of diabetes in the smart Internet world, there is a need for an automated prediction system to facilitate the patients, to get know, whether they are affected by the disease or not. There are many diabetes prediction software that is already in use, still, the accurateness of a diabetes prediction is not complete. This paper presents a robust framework (PSO-NNDP), employs a novel hybrid feature selector to improvise the neural-based diabetes prediction system. The novel hybrid feature selector presented in this paper comprises the merits of the correlation coefficient, F-score, and particle swarm optimization methods to influence the feature selection process. The reliability of the proposed framework has been experimented on the benchmarking dataset. By establishing the clear steps, for the replacement of missing values, removal of outliers, the proposed framework obtains 99.5% accuracy. Moreover, the experimented machine learning models also show a great improvement upon the usage of the proposed feature selector.