Emotion Recognition Using Multifractal Features of Electroencephalogram Signals and Machine Learning Methods

Abstract

Background: Emotion recognition based on electroencephalogram (EEG) signals plays a significant role in a wide range of applications, such as brain-computer interfaces and intelligent interactive systems. The present study investigates the impact of multifractal features derived from EEG signals on the accurate and automatic detection of emotional dimensions using artificial intelligence-based models. Methods: EEG data were obtained from the labeled DREAMER dataset, consisting of 14 channels recorded from 23 healthy participants. Following preprocessing steps and decomposition into alpha, beta, and gamma frequency bands, multifractal features of the EEG signals were extracted using the Multifractal Detrended Fluctuation Analysis (MFDFA) method. Classification of the emotional dimensions—arousal, valence, and dominance—was performed using the Support Vector Machine (SVM) algorithm. Findings: Multifractal features extracted from the alpha band exhibited the highest discriminative power among emotional states compared to other frequency bands. These features not only simplified the computational process but also improved the accuracy of emotion recognition in the arousal, valence, and dominance dimensions to over 96, 97, and 93, respectively. Conclusion: Multifractal features—particularly in the alpha band—are highly effective in extracting emotional information from EEG signals. The proposed EEG-based feature extraction method can accurately distinguish between different emotional dimensions and can be utilized for the development of intelligent neuro-interactive systems. © 2026, Isfahan University of Medical Sciences(IUMS). All rights reserved.