Developing a Framework for Human Emotion Detection and Stress Analysis using Biomedical Signals

Abstract

In recent years, advances in machine learning techniques have opened up new options for automated recognition of emotional and stress states using electroencephalogram (EEG) signals. This work describes a thorough machine-learning approach to emotion and stress detection, which uses EEG data to create an accurate and dependable recognition system. The suggested methodology is a multi-step procedure that begins with the capture and preprocessing of EEG data, followed by feature extraction using wavelet transform and statistical methods. These features are then fed into a variety of machine learning classifiers, such as Support Vector Machines (SVM), Neural Networks (NN), and Decision Trees (DT), which discover patterns associated with distinct emotional and stress states.

Filtering to reduce noise and artefacts is part of the preprocessing phase, which ensures that the data used for the subsequent analysis is clean and meaningful. Feature extraction aims to capture both time-domain and frequency-domain properties of EEG data, which are essential for discriminating various mental states. The classifiers are trained and verified on a labelled dataset, allowing the system to learn and generalise patterns associated with different emotions and stress levels.

Our experimental results show that the proposed technique is highly accurate in recognizing emotional and stress states. The SVM classifier performs very well, with an accuracy of 92%, followed by Neural Networks at 89% and Decision Trees at 85%. These classifiers' performance is measured using metrics including accuracy, precision, recall, and the area under the Receiver Operating Characteristic (ROC) curve (AUC).

To demonstrate the success of our approach, we compared the ROC curves for filtered and unfiltered EEG data. The ROC curve for filtered data has a steeper rise and a higher AUC, indicating greater discriminative power and fewer false positives than unprocessed data. This emphasizes the relevance of preprocessing in improving the performance of classification models.

This study highlights the potential of applying machine learning approaches to automate emotion and stress assessment using EEG signals. The findings show that our suggested approach is a reliable tool for real-time mental health monitoring, with important implications for applications in healthcare, workplace stress management, and human-computer interaction. Future research will concentrate on integrating real-time EEG recording devices and investigating deep learning models to enhance detection accuracy and robustness.