The objective of this research is to comparatively study the analysis of depression potentiality from videos recorded during PHQ-9 assessment and clinical psychologists’ expertise (psychometric) in major depression disorder diagnosis by implementing experimental research from three subject groups. The subjects were selected by using the purposive sampling method, and they consented according to the Human Research Ethics Committee. In addition, the data were collected via experimentation at the Psychiatry and Drug Addiction Subdivision of the hospital by using a research tool created from the application of facial expression recognition pre-training model, user behavior tracking, and PHQ-9 questionnaire in accordance with clinical manifestations. The relationship between PHQ-9 questions and the subjects’ emotions, which occurred while they are taking the questionnaires, are analyzed by using correlation statistics. After that, the results are compared by three clinical psychologists who have been specializing in depression diagnosis for more than seven years. The relationship exhibits the following interesting outcomes: 1) The correlation result showed a significant relationship between PHQ-9, emotion, behavior tracking, and reaction time with real subjects (Experimental group) but not significant with fake subjects (Experimental excluded group); and 2) Most of the emotions derived from the experiment have similar trends. To elaborate, the untreated subjects and the subjects who are undergoing treatment are prone to sadness, whereas the normal subjects tend to be happy, and the result barely has a relationship with PHQ-9

This research compares the capability of data classification models to predict consistent results for a subject’s depression potentiality, track the subject behaviour and recognise facial expressions during PHQ-9 assessments. This research is motivated by the necessity for depression screening and diagnosis, which traditionally relies on observations by experienced physicians or clinical psychologists of symptoms in conjunction with data from questionnaires. However, the field still requires a suitable technological approach that gives more accurate and consistent results. All data used in the present research were collected by combining technologies and compared by using classification models, the goal being to find the machine-learning model that most accurately predicts consistent results for the subjects’ PHQ-9 assessment, behaviours and emotions. The subjects were screened by clinical psychologists and divided into three groups: (i) subjects suffering from depression but not receiving treatment (undertreated subjects), (ii) subjects undergoing depression treatment (subjects undergoing treatment) and (iii) subjects without depression disorder (normal subjects). Related studies have compared the accuracy of classification models to one another. The four most frequently applied classification models in depression-related studies are (i) decision tree (ii) support vector machine, (iii) naïve Bayes and (iv) neural network. All models were analysed, designed and developed before being tested experimentally. The accuracy of the experimental results was tested by using the data analysis tool RapidMiner Studio. The results show that the decision tree model is not only the most accurate for predicting depression potentiality, tracking behaviour and recognising facial expressions during PHQ-9 assessments but also the most suitable.