Multiscale Fractal Analysis on EEG Signals for Music-Induced Emotion Recognition

TL;DR

This paper introduces novel multiscale fractal analysis algorithms for EEG signals to improve emotion recognition, demonstrating superior performance over traditional features on the DEAP dataset.

Contribution

It presents two new algorithms based on Multiscale Fractal Dimension and Multifractal Detrended Fluctuation Analysis for extracting emotion-related features from EEG signals.

Findings

Outperforms baseline features on DEAP dataset

Multifractal analysis enhances robustness of emotion recognition

Proposes efficient feature extraction methods for EEG analysis

Abstract

Emotion Recognition from EEG signals has long been researched as it can assist numerous medical and rehabilitative applications. However, their complex and noisy structure has proven to be a serious barrier for traditional modeling methods. In this paper, we employ multifractal analysis to examine the behavior of EEG signals in terms of presence of fluctuations and the degree of fragmentation along their major frequency bands, for the task of emotion recognition. In order to extract emotion-related features we utilize two novel algorithms for EEG analysis, based on Multiscale Fractal Dimension and Multifractal Detrended Fluctuation Analysis. The proposed feature extraction methods perform efficiently, surpassing some widely used baseline features on the competitive DEAP dataset, indicating that multifractal analysis could serve as basis for the development of robust models for affective state recognition.