Automated Detection of Diabetes by Means of Higher Order Spectral Features Obtained from Heart Rate Signals

Diabetes Mellitus is a chronic disease that affects a large population around the world. Electrocardiogram (ECG), can assist in diagnosing DM by identifying abnormalities within the signals. In this study, we have developed an algorithm that uses Heart Rate Variability (HRV) obtained from ECG signals to detect diabetes. Third order cumulants of Higher Order Spectra (HOS) is used to analyze the HRV signals. Principle Component Analysis (PCA) and student test is employed to extract features and select them before classifying them using seven different classifiers, namely Fuzzy classifier, Gaussian Mixture Model (GMM), Support Vector Machine (SVM), Probabilistic Neural Network (PNN), Naive Bayes classifier (NBC), K-Nearest Neighbour (KNN) and Decision Tree (DT) classifier. The performances of the classifiers were evaluated by calculating values of their accuracy, Positive Predictive Value (PPV), sensitivity and specificity. The accuracy for the classifiers ranges between 52.76% and 79.93%. SVM using RBF kernel has achieved the highest accuracy with 79.93%.