Emotion Detection and Classification Using Machine Learning Techniques

Abstract

This chapter analyzes 57 articles published from 2012 on emotion classification using bio signals such as ECG and GSR. This study would be valuable for future researchers to gain an insight into the emotion model, emotion elicitation and self-assessment techniques, physiological signals, pre-processing methods, feature extraction, and machine learning techniques utilized by the different researchers. Most investigators have used openly available databases, and some have created their datasets. The studies have considered the participants from the healthy age group and of similar cultural backgrounds. Fusion of the ECG and GSR parameters can help to improve classification accuracy. Additionally, handcrafted features fused with automatically extracted deep machine learning features can increase classification accuracy. Deep learning techniques and feature fusion techniques have improved classification accuracy.

Multimodal and Multidomain Feature Fusion for Emotion Classification Based on Electrocardiogram and Galvanic Skin Response Signals

“…An outline of the emotion classification accuracies reported by the researchers using machine learning techniques is mentioned below. Egger [13]. The DEAP database provided physiological signals for emotional measurements for conducting research [15].…”

“…While researchers have explored using EEG signals for emotion classification, this method is more suitable for clinical applications. ECG and GSR signals have been used less frequently for emotion classification compared to EEG signals [13]. An ECG records the heart's electrical movement, while the GSR measures the skin's electrical conductance.…”

Emotion Classification Based on CWT of ECG and GSR Signals Using Various CNN Models

“…Using a CNN-based deep learning approach with automatic feature extraction simplifies the classification task. Dessai et al analyzed studies on emotion classification using ECG and GSR signals [20]. They concluded that deep learning techniques using automatic extraction of features increased the classification accuracy compared to traditional machine learning techniques.…”

“…Deep learning models such as CNN, long short-term memory (LSTM), and recurrent neural network (RNN) are also used for classifying emotions [20]. Al Machot et al obtained classification accuracy of 78% using the MAHNOB database and 82% accuracy using the DEAP database on GSR data using a CNN classifier [14].…”

“…Figure 3 indicates the ECG signal waveform [12]. GSR signals were captured using the Shimmer 2R with a sampling frequency of 128 Hz and 12-bit resolution, with two electrodes placed at the middle of the left middle and index fingers, as indicated in Figure 4 [11,20]. GSR signals were captured using the Shimmer 2R with a sampling frequency of 128 Hz and 12-bit resolution, with two electrodes placed at the middle of the left middle and index fingers, as indicated in Figure 4 [11,20].…”