Enhanced Pre-processing Strategies for Accurate Diabetes Prediction in Healthcare using Noval Method: ANN+LDA

Abstract

In the recent past, so many chronic diseases have been emerging and spreading in the world and even in the developing and advanced countries as well. One of such serious chronic diseases is Diabetes Mellitus that covers and impacts the health of people from early age. Nevertheless, the available Machine Learning (ML) and Deep Learning (DL) approaches are unable to provide good predictions in patients relating to diabetes. In addition, this study evaluated the proposed pre-processing procedure on large datasets for diabetes prediction that contained outliner detection and removal, missing values imputation, and standardization, to improve diabetes ascertainment. This research evaluated the proposed pre-processing procedure on a large set of data by outlier identification and removal, missing values imputation and data standardization were done to improve diabetes forecast. To ensure rapid and accurate classification of diabetes, the researchers employed and initialized an Artificial Neural Network (ANN). Data was gathered from the PIMA Dataset and North California State University (NCSU). Following this, Bivariate filter was applied to sort out features which were relevant. The selected features were subsequently subjected to Pearson correlation towards feature set refinement considering a threshold below which features were eliminated and only the most effective features selected. From the results it was evident that the proposed approach was significantly better than the existing methods in terms of accuracy as it achieved a classification accuracy of around 93% as opposed to the other methods.