An Ensemble Learning Approach for Electrocardiogram Sensor Based Human Emotion Recognition.

Area of Science:

• Computational intelligence and machine learning applied to affective computing.
• Biomedical signal processing and biosensor technology for human-computer interaction.

Background:

• Current biosensor-based human emotion recognition models often struggle with high classification accuracy, especially when using limited biosensors.
• Ensemble learning methods have demonstrated success in improving classification performance across various machine learning tasks.

Purpose of the Study:

• To develop an ensemble learning model for recognizing four major human emotions (anger, sadness, joy, pleasure) using electrocardiogram (ECG) signals.
• To enhance emotion classification accuracy by combining novel and established ECG feature extraction techniques with ensemble learning and feature selection.

Main Methods:

• Feature extraction from ECG signals using heart rate variability, empirical mode decomposition, within-beat analysis, and a novel frequency spectrum analysis.
• Implementation and evaluation of various ensemble learning algorithms for emotion classification.
• Application of feature selection prior to ensemble model training to optimize performance.

Main Results:

• The developed ensemble learning model achieved a 10.77% accuracy gain over the best-performing single biosensor model.
• The proposed model demonstrated superior classification accuracy compared to most existing multi-biosensor emotion recognition systems.

Conclusions:

• Ensemble learning, combined with advanced ECG feature extraction and selection, offers a powerful approach for accurate human emotion recognition.
• This method presents a significant advancement in affective computing, potentially enabling more sophisticated human-computer interactions.