Detection of Mental Stress in Sports University Students through Machine Learning Techniques

Abstract

Mental stress is a significant issue affecting young individuals, particularly students, and can lead to serious cognitive disabilities, if left unaddressed. This study aimed to develop a machine-learning-based system to detect and measure stress levels in university-level sports students using vocal and acoustic features. Data were collected from 2400 students at the Lakshmibai National Institute of Physical Education (NERC), Guwahati, and analysed using Convolutional Neural Network (CNN) and random forest (RF) classification algorithms. The impact of exam pressure, match pressure, and recruitment stress on mental stress levels was examined. The performance of the algorithms was evaluated using the accuracy, precision, recall, and F1-score metrics. The RF algorithm achieved the highest accuracy (91.1 %) among the two classifiers. The proposed system aims to provide an objective tool for assessing stress levels, enabling earlier intervention, and more effective management of stress-related conditions by clinical psychologists. The study hypothesized that certain vocal characteristics, such as pitch variability, energy, Mel Frequency Cepstral Coefficients (MFCCs), Linear Predictive Coding Coefficients (LPCC), Zero Crossing Rate (ZCR), formant extraction, tempo beat extraction, and tonnetz extraction, would exhibit a significant correlation with higher stress levels. This study reviewed research on stress detection with machine learning, summarizing methods and classifier performance.