Augmenting ECG Data with Multiple Filters for a Better Emotion Recognition System

Abstract

A physiological-based emotion recognition system (ERS) with a unimodal approach such as an electrocardiogram (ECG) is not as popular compared to a multimodal approach. However, a single modality has the advantage of lower development and computational cost. Therefore, this study focuses on a unimodal ECG-based ERS. The ECG-based ERS has the potential to become the next mass-adopted consumer application due to the wide availability of wearable and mobile ECG devices in the market. Currently, ECG-inclusive affective datasets are limited, and many of the existing datasets have small sample sizes. Hence, ECG-based ERS studies are stunted by the lack of quality data. A novel multi-filtering augmentation technique is proposed here to increase the sample size of the ECG data. This technique augments the ECG signals by cleaning the data in different ways. Three small ECG datasets labelled according to emotion state are used in this study. The benefit of the proposed augmentation techniques is measured using the classification accuracy of five machine learning algorithms; k-nearest neighbours (KNN), support vector machine, decision tree, random forest and multilayer perceptron. The results show that with the proposed technique, there is a significant improvement in performance for all the datasets and classifiers. KNN classifier improved the most with the augmented data and the reported classification accuracies of over 90%.

Smart Healthcare: Exploring the Internet of Medical Things with Ambient Intelligence

“…These issues can be mitigated through filtering techniques such as the Butterworth filter, which operates within a frequency range of 0.5-8 Hz [25]. In CBGM, noise removal approaches include adaptive iterative filtering and fast discrete lifting-based wavelet transform (LWT) [26] as well as multi-filtering augmentation [27]. Pulse oximetry sensor data often utilize adaptive filtering techniques [28], while accelerometer and gyroscope sensors benefit from Butterworth high-pass filtering [29], complementary filters, and Kalman filters [30] for error assessment and enhanced accuracy.…”

Review of Studies on Emotion Recognition and Judgment Based on Physiological Signals

“…They pointed out that importance weighting in machine learning models could reduce the effects of individual physiological differences in peripheral physiological responses. Hasnul et al [18] presented a new multi-filtering augmentation algorithm to increase the sample size of the ECG data. The algorithm augmented ECG signals by cleaning the data in different ways, and the benefit of the algorithm was measured using the classification accuracy of five machine learning algorithms.…”

Asymmetric Windowing Recurrence Plots on Input Formulation for Human Emotion Recognition

“…While research on emotion recognition has spanned various modalities such as facial expressions [5], [6], gestures [7], [8], voice styles [9]- [11], physiological signals such as electrocardiography (ECG) [12], [13], galvanic skin response (GSR) [14], [15], and EEG have emerged as more reliable modalities. EEG is particularly noteworthy because of its intrinsic linkage with the activity of the central nervous system, making it resilient to falsification and manipulation [16].…”