Multi-Class Classification of Abnormal Heartbeat Detection using Hyperdimensional Computing

Hyperdimensional computing (HDC) is a brain-inspired computing method that is applicable to a wide range of classification tasks. This paper addresses group-specific multi-class beat detection using the HDC method with data from the MIT-BIH arrhythmia and St. Petersburg INCART 12-Lead arrhythmia database. Four different types of features are extracted from the data. Minimum redundancy maximum relevance (mRMR) and the feature selection method based on Area under the receiver operating characteristic (ROC) Curve (AUC) value, called AUC ranking feature selection, are used to select features for classification. A concatenated approach is employed to convert features into hyperdimensional space. The HDC classifier with the AUC ranking feature selection achieves a higher AUC value and better performance compared to the mRMR feature selection method in the One-vs-Rest (OVR) multi-class classification. It demonstrates that the AUC ranking feature selection method achieves better accuracy performance compared to the mRMR feature selection method. Based on the AUC ranking feature selection method, HDC approach attains a test accuracy exceeding 80% in the One-vs-One (OVO) classification task using seed hypervectors of low dimensionality, i.e., with dimensionality of 100 bits only. Also HDC classifier maintains stable performance in comparison to support vector machine (SVM) and multilayer perceptron (MLP) models.