Systematic literature review of mental stress recognition using wearable sensor data fusion

Abstract

Mental stress is a widespread issue in modern society, significantly impacting individuals' well-being and productivity across various demographic groups. Detecting and managing mental stress is crucial to addressing its adverse effects on physical and psychological health. Traditional methods rely on subjective assessments, which may lack accuracy and scalability. This paper presents a systematic review exploring the use of methods that combine information (Fusion Techniques) from various wearable sensors to mental stress recognition by using machine learning algorithms.
The focus was on identifying trends in classifiers, data fusion techniques, sensors, and evaluation metrics. The findings highlight Support Vector Machine (SVM) as the most effective classifier, followed by Random Forest (RF) and K-nearest Neighbors (KNN). ECG (Electrocardiogram) and EEG (Electroencephalogram) emerged as the most used sensors due to their ability to monitor cardiovascular and brain activity. Metrics such as accuracy, precision, and F1 score were predominant in evaluating model performance.
The review reveals a strong preference for aggregating features extracted from diverse raw data sources which enhances robustness of mental stress detection by using machine learning algorithms. While existing studies demonstrate significant advancements, the findings indicate opportunities for further improvement in hybrid fusion techniques and real-world applications.