Irregular heartbeat detection using sequentially truncated multilinear singular value decomposition

This paper presents a novel approach for detecting irregular heartbeats using tensor approximation and Least Squares — Support Vector Machines (LS-SVM). Once the signal is filtered and normalized, a third order tensor was constructed for each record of the dataset. Next, a Sequentially Truncated Multilinear Singular Value Decomposition (ST-MLSVD) was applied and the mode-3 matrix was used as input features for an LS-SVM. Then, Active Prototype Vector (APV) selection strategy was performed for selecting 5% of the data for training. The LS-SVM hyper-parameters tuning was carried out using a combination of Coupled Simulated Annealing and the simplex method. Two databases were used for the performance evaluation. This evaluation resulted in sensitivities, positive predictive values and specificities all above 93%. These results are an improvement on previously reported results for tensor-based irregular heartbeat detection systems.