Unveiling the impact of dataset size on machine learning models for anxiety and depression prediction amid the COVID-19 pandemic: determining optimal data collection thresholds

Our emotional, psychological, and social well-being are all parts of our mental health. An individual's routine can be disrupted and their mental is health affected by stress, despair, and anxiety. Mental health preservation and restoration are essential for each person as well as for communities and society as a whole. The COVID-19 pandemic has triggered a strong emotional and psychological reaction in many people, in addition to triggering a global health emergency. The pandemic's uncertainty, disruptions, and social changes have amplified stress, fear, and depression, which are common responses to crises. Data collection for the COVID-19-related depression and anxiety assessment was limited to online methods because of the ongoing pandemic. In the field of mental health evaluation, the application of machine learning techniques has emerged as a promising strategy for identifying and grasping anxiety and depression symptoms. This paper employed K-Nearest Neighbors (kNN), Random Forest (RF), Decision Tree (DT), and Support Vector Machine (SVM) techniques on the prevalence of anxiety and depression among Bangladeshi university students during the COVID-19 pandemic. This paper addresses how the accuracy of predictions made by various machine learning models is affected by the size of the datasets. The findings of this study illuminate the scalability and generalizability of different machine-learning methods. The findings validate that how accuracy of the models has consistently and significantly improved as the dataset size varies. The performance of classification models is further assessed using the F1 score, precision, and recall.