A Novel Exploitative and Explorative GWO-SVM Algorithm for Smart Emotion Recognition

TL;DR

This paper introduces a new ECG-based emotion recognition method using an optimized SVM algorithm, demonstrating improved reliability and efficiency suitable for real-time embedded systems, outperforming previous machine learning approaches.

Contribution

The paper proposes the X-GWO-SVM algorithm for ECG-based emotion recognition, combining exploitative and explorative GWO optimization to enhance accuracy and efficiency over existing methods.

Findings

X-GWO-SVM outperforms other supervised machine learning methods in reliability.

The algorithm is suitable for implementation in lightweight embedded systems.

Demonstrates superior performance on both self-collected and benchmark datasets.

Abstract

Emotion recognition or detection is broadly utilized in patient-doctor interactions for diseases such as schizophrenia and autism and the most typical techniques are speech detection and facial recognition. However, features extracted from these behavior-based emotion recognitions are not reliable since humans can disguise their emotions. Recording voices or tracking facial expressions for a long term is also not efficient. Therefore, our aim is to find a reliable and efficient emotion recognition scheme, which can be used for non-behavior-based emotion recognition in real-time. This can be solved by implementing a single-channel electrocardiogram (ECG) based emotion recognition scheme in a lightweight embedded system. However, existing schemes have relatively low accuracy. Therefore, we propose a reliable and efficient emotion recognition scheme - exploitative and explorative grey wolf optimizer based SVM (X - GWO - SVM) for ECG-based emotion recognition. Two datasets, one raw self-collected iRealcare dataset, and the widely-used benchmark WESAD dataset are used in the X - GWO - SVM algorithm for emotion recognition. This work demonstrates that the X - GWO - SVM algorithm can be used for emotion recognition and the algorithm exhibits superior performance in reliability compared to the use of other supervised machine learning methods in earlier works. It can be implemented in a lightweight embedded system, which is much more efficient than existing solutions based on deep neural networks.