Analysis of Schoolgirls' Mental Health – A Machine Learning Approach to Distinguish Between Academic and Abusive Stress

Abstract

The prevalence of stress and depression is rising among students. Current research acknowledges that academic burden is a source of stress for students and evaluates physiological parameters during academic activities by deliberately introducing stressors or stress. Nonetheless, schoolgirls perceive sexual, physical, and emotional abuse as the primary source of stress nowadays. This study gathered information from fifty schoolgirls via an in-house oral questionnaire to identify those concerned about academic pressure and maltreatment. This questionnaire was designed to disclose the psychological state through oral, behavioural, and physiological responses without instigating stress unnaturally. The prepared dataset was experimented with using multinomial logistic regression and decision tree using C4.8, Naive Bayes, and k-NN for three class classifications (normal, academic, and abuse stress). The weighted average F1-score of the leading models from each classifier was 89.9%, 89.2%, 89.5%, and 88.6%, respectively. The results indicate that logistic regression performs marginally better than other methods. When separating normal and academic stress samples, the same classifier achieves an F1-score of 94.9% and 69.0%, respectively. In identifying the abuse stress classes, k-NN achieved the maximum F1-score of 88.8%. In addition, the normal, academic, and abuse stress classes achieved sensitivities of 98.4%, 65.4%, and 98.4% and specificities of 71.7%, 98.4%, and 100%, respectively. The categorisation models constructed from the research can identify schoolgirls with internalised conflict for earlier intervention.