Interpretable Dual-Filter Fuzzy Neural Networks for Affective Brain-Computer Interfaces

TL;DR

The paper addresses the challenge of interpretable emotion decoding in affective brain-computer interfaces by introducing iFuzzyAffectDuo, a dual-filter fuzzy rule-based model with spatial and temporal filters and Modified-Laplace membership functions. The approach emphasizes explainability by mapping neural signals to human-understandable fuzzy rules while maintaining high accuracy. Evaluations on two fNIRS datasets and one EEG dataset show state-of-the-art performance, with clear interpretability insights into neural patterns and channel contributions. The work advances affective computing and neuroscience by providing a robust, interpretable framework that can be extended to additional modalities such as fMRI and ECoG and adapted for online use.

Abstract

Fuzzy logic provides a robust framework for enhancing explainability, particularly in domains requiring the interpretation of complex and ambiguous signals, such as brain-computer interface (BCI) systems. Despite significant advances in deep learning, interpreting human emotions remains a formidable challenge. In this work, we present iFuzzyAffectDuo, a novel computational model that integrates a dual-filter fuzzy neural network architecture for improved detection and interpretation of emotional states from neuroimaging data. The model introduces a new membership function (MF) based on the Laplace distribution, achieving superior accuracy and interpretability compared to traditional approaches. By refining the extraction of neural signals associated with specific emotions, iFuzzyAffectDuo offers a human-understandable framework that unravels the underlying decision-making processes. We validate our approach across three neuroimaging datasets using functional Near-Infrared Spectroscopy (fNIRS) and Electroencephalography (EEG), demonstrating its potential to advance affective computing. These findings open new pathways for understanding the neural basis of emotions and their application in enhancing human-computer interaction.

Figures (5)

• Figure 1: Overview of the iFuzzyAffectDuo model and related components.(A) Example of the fNIRS signal recorded during exposure to affective stimuli, showcasing both negative and neutral responses. (B) Architecture of the fuzzy filter employed in the model. (C) Brain orientation for all 3D plots used in the analysis. (D) Structural layout of the proposed iFuzzyAffectDuo model. (E) Center prototypes of the spatial filter. (F) Firing Strengths of the spatial filter. (G) Center prototypes of the temporal filter. (H) Firing Strengths of the temporal filter.
• Figure 2: Modified-Laplace MFs $m$ with different center parameters $m$ and $\lambda$
• Figure 3: Comparative performance of the iFuzzyAffectDuo model against DBJNet, EEGNET, NLSTM, and Transformer models across three datasets. This figure demonstrates the model's robust performance superiority across diverse machine learning challenges. ***$p < 0.001$
• Figure 4: Dynamic Distribution of Fuzzy Set Membership and Feature Distributions Across the FACED Dataset at Key Time Intervals. This figure illustrates the variation in fuzzy membership degrees and overlays of neural response histograms. The histograms represent the EEG data in query space $W_r^Q$, while the colored curves denote different Modified-Laplace MFs applied to decode underlying neural mechanisms.
• Figure 5: Sample analysis: Visualization of membership degrees and firing strengths in rule #1.