A Supervised Learning Approach to Predicting Coronary Heart Disease Complications in Type 2 Diabetes Mellitus Patients

Abstract

A supervised machine learning approach that incorporates genetic algorithms (GA) and weighted k-nearest neighbours (WkNN) was applied to classify type 2 diabetes mellitus (T2DM) patients according to the presence or absence of coronary heart disease (CHD) complications. The investigation was carried out by analyzing potential risk factors recorded at the Ulster Hospital in Northern Ireland. A GA initialization technique that integrates medical expert knowledge was compared with traditional data-driven GA initialization techniques. The results indicate that the incorporation of expert knowledge provides only a small improvement of CHD classification performance compared with models based on data-driven initialization techniques. This may be due to data incompleteness and noise or due to the beneficial effects of treatment, which masks the complication of CHD in the dataset. Further incorporation of expert knowledge at different levels of the GA need to be addressed to improve decision support in this domain