Identification of type-2 diabetes by electrocardiogram signal using flexible analytical wavelet transform

Type-2 diabetes mellitus (T2DM) is a lifelong metabolic disease with worldwide prevalence. It can drastically decrease the life expectancy of any subject with a huge economic burden. The present study aimed to create a non-invasive and economical tool for automatic detection of T2DM using electrocardiogram (ECG) signals. The flexible analytic wavelet transform is used to evaluate the ECG by decomposing it into predictable sub-bands. Statistical and time-domain features were extracted from each sub-band. Different feature selection techniques were applied to obtain the most relevant features. The top nine features, selected by using the one-R attribute eval feature selection technique, were fed into the various types of machine learning classifiers. In tested classifiers, the fine k-nearest neighbour and optimisable KNN classifiers have shown the highest average accuracy of 94.94% and 94.61% respectively. The results suggest that the proposed approach provides an efficient non-invasive T2DM detection method in regular applications.