Machine learning techniques to classify emotions from electroencephalogram topographic maps: A systematic review

• Published in: Computers in biology and medicine Vol. 197; no. Pt A; p. 111022
• Main Authors: Melo, Marla P., Adamatti, Diana F., Aguiar, Marilton S.
• Format: Journal Article
• Language: English
• Published: United States Elsevier Ltd 01.10.2025
• Subjects: EEG topographic map; Electroencephalography - methods; Emotions - physiology; Humans; Internal Medicine; Machine Learning; Neural Networks, Computer; Other; Recognition emotions; Signal Processing, Computer-Assisted; EEG topographic map; Machine learning; Recognition emotions
• ISSN: 0010-4825; 1879-0534; 1879-0534
• DOI: 10.1016/j.compbiomed.2025.111022

In the task of image classification for emotion recognition, facial expression data is commonly used. However, electrical brain signals generated by neural activity provide data with greater integrity. We can capture these signals non-invasively using electroencephalogram (EEG) recording devices. Conversion techniques can graphically represent the signal information as EEG topographic maps (ETMs). This review aims to identify machine learning techniques for recognizing emotional states from EEG topographic data maps. This review follows the PRISMA guidelines, and we conducted the literature search up to July 2025. Fourteen publications met the inclusion criteria. The identified machine learning techniques encompass a range of models, from Support Vector Machines (SVM) to deep neural models, which include seven Convolutional Neural Networks (CNNs), a lightweight convolutional neural network (LCNN), a Visual Geometry Group network (VGG-16), two Bidirectional Long Short-Term Memory networks (Bi-LSTM), two Residual Networks (ResNet), and a Multilayer Perceptron (MLP). This review presents the state of the art by providing a comprehensive mapping of machine learning techniques used for emotion recognition based on EEG topographic maps. It also summarizes the correlations evaluated in the fourteen studies, including emotional datasets, feature extraction techniques, and approaches for converting EEG signals into EEG topographic maps. Furthermore, it discusses classification accuracy based on subject-dependent, subject-independent, transfer learning, and cross-subject approaches, offering insights into potential directions for future research. •It presents an overview of emotion recognition based on image-converted EEG signals.•It presents methods for converting electroencephalogram signals into images.•Provides a mapping of feature extraction techniques for electroencephalogram signals.•Correlates factors that influence the improvement of emotion recognition models.•Guidance for future research in machine learning on electroencephalogram signal data.