A Neural Network based Diabetes Prediction on Imbalanced Data

As an extensively well-known chronic disease, diabetes is an illness that harms the body's capability to process blood glucose. The proper treatment of diabetes could help a person live a long and normal life in general. It is necessary to detect the disease at an early stage. We focus our work on the performance of a machine-learning (ML) algorithm to identify the presence of diabetes on the PIMA Indian diabetes dataset (PIDD) which referenced from the University of California, Irvine (UCI) ML repository. Using ML, we know about the classification and prediction techniques. Further, diabetes became an attention seeker in the field of research due to the presence of imbalanced and missing data. Although many factors affect the performance of the algorithm, This research paper worked on the prediction technique for diabetes classification with outliers and missing values in data with class imbalance. Using an adaptive synthetic sampling method (ADASYN) and reduced the impact of class imbalance on the performance of the prediction model. Then, this algorithm improved the generalization using a feature selection technique and multilayer perceptron classifiers to make predictions and evaluations. Experimental results shows that this experiment obtained a better accuracy of 84% with a neural network model in comparison with the previous model.