A novel approach for the identification of chronic alcohol users from ECG signals

Several medical studies reveal that alcohol consumption has pronounced effects on the heart rate variability (HRV) of the consumer. In this paper, machine learning algorithms use the features extracted through HRV analysis performed on ECG samples of chronic alcohol users and normative subjects, in order to classify them. To carry out the classification, a Support Vector Machine (SVM) and an Extreme Learning Machine (ELM) have been trained, and their performance has been compared. While time domain, frequency domain and non-linear features are generally extracted from ECG signals for HRV analysis, in this study a new set of features obtained from Autoregressive Modelling using Exogenous (ARX) Inputs have also been used to improve the accuracy of the algorithms. An accuracy of 80.36% and 89.29% was achieved by SVM and ELM respectively without the use of ARX coefficients, while an accuracy of 87.50% and 94.64% was achieved when ARX coefficients were included in the feature set. The use of ARX coefficients and the ELM classifier is the novelty of this paper, which helped in increasing the accuracy of the classification system.