XAI_MLPCNN: A Novel Explainable AI-Based DeepLearning Framework for Stress Identifi cation

Abstract

Over the past ten years, there has been a lot of emphasis focusedon the development of Artificial Intelligence (AI) and Machine Learning(ML)-based mental health treatments. To increase practitioners' and patients'trust in AI applications, AI systems need to explain their actions. This iscalled Explainable AI (XAI). While significant progress has been achievedin stress prediction models, XAI has not advanced as much. To overcomethis gap, this work presents an explainable AI-based Multi-Layer PyramidConvolutional Neural Network (XAI_MLPCNN) architecture for stressdetection. Multi-channel EEG recordings can be deconstructed into distinctfrequency bands and their non-linearity and non-stationarity removed usingthe Discrete Wavelet Transform (DWT). When processing features, thePower Spectral Density (PSD) is employed. Conversely, the decomposedsignals are employed in the automatic feature extraction process throughMLPCNN, and the dual BiLSTM with self-attention layer (DBiL_SA) isutilized to predict stress. MLPCNN-DBiL_SA and PSD features arecombined to improve prediction. To provide explanations or assess howexplainable the predictions are, explainable artificial intelligence techniqueslike Shapley additive explanations (SHAP) and Local Interpretable Model-agnostic Explanations (LIME) are employed. The Python platform is used toimplement the model. Performance is further assessed using a severalperformance metrics, such as accuracy, recall, precision, and f1-measure.Furthermore, the proposed approach is compared to other methods that arecurrently in use, like CNN-DWD and PSD, LSTM-DWD and PSD, BI-LSTM-DWD and PSD, RNN-DWD and PSD, and GRU-DWT AND PSD.