PERFORMANCE AND UNCERTAINTY EVALUATION IN MACHINE LEARNING MODELS FOR PERSONALITY CLASSIFICATION

Abstract

The classification of personality types is essential in mental health screening, career counseling, and more specific interventions in health care in line with the UNSDG 3 (Good Health and Well-Being). This paper compares machine learning methods for classifying personality types (Introvert, Extrovert, Ambivert) based on a dataset of 20,000 samples and 31 behavioral and psychological variables. Eight models in total were considered as part of robust preprocessing (removing outliers, selecting features, and having a stratified train-test partition) and considering a new uncertainty quantification framework focusing on predictive, aleatoric, and epistemic uncertainties. The Support Vector Machine (SVM) performed best, with accuracy, precision, recall, and F1 score of 0.990786, 0.990794, 0.990786, and 0.990789, respectively, and minimum values of uncertainty (predictive: 0.0248067 and epistemic: 0.00266215). These findings suggest the potential application of SVM in mental health screening, along with individual well-being interventions, in line with the goals of UN Sustainable Development Goal 3. The methodology we have developed helps us address fundamental uncertainties in the data and is a stronger method for reliable personality classification.