Original Data Vs High Performance Augmented Data for ANN Prediction of Glycemic Status in Diabetes Patients

In the following article a comparative analysis between Original Data (OD) and Augmented Data (AD) are carried out for the prediction of glycemic status in patients with diabetes. Specifically, the OD concerning the time series of the glycemic status of a patient are compared with AD. The AD are obtained by the randomised average with five different ranges, and are processed by a Machine Learning (ML) algorithm for prediction. The adopted ML algorithm is the Artificial Neural Network (ANN) Multilayer Perceptron (MLP). In order to optimise the prediction two different data partitioning scenarios selecting training datasets are analysed. The results show that the algorithm performances related to the use of AD through the randomisation of data in different ranges around the average value, are better than the OD data processing about the minimization of statistical errors in self learning models. The best achieved error decrease is of 75.4% if compared with ANN-MLP processing of the original dataset. Furthermore, in the paper is added a linked discussion about the economic and managerial impact of AD in the healthcare sector.