From Personality Prediction to Behavioral Subtyping: A Machine Learning Framework for Mental Health Assessment

Abstract

Introversion and extroversion are personality traits that shape individuals behaviours; they equally influence habits, attitude, and cognition. Modern psychological theories define the introvert-extrovert spectrum as a continuous dimension of personality, rather than two distinct personality types. It is essential to recognize this continuity in order to better understand and support individual differences,and it remains difficult to capture these nuances using computational methods, as most existing studies rely on textual data or data from questionnaires, which can introduce bias. In our study, we investigate personality classification using behavioral features, as behavioral data offer an unexploited path to understanding personality beyond self-reported questionnaires, by leveraging a dataset  that includes key indicators for 2,900 individuals and 08 behavioral features. Our methodology integrates both supervised and unsupervised learning to address complementary objectives, where supervised classifiers are used to establish a predictive framework for the extraversion-introversion dichotomy, while unsupervised clustering is used to test whether the resulting latent structure corresponds to these binary labels. Through combining predictive and exploratory methods, our work contributes both to benchmarking behavioral personality classification, providing evidence  that certain prediction errors may correspond to significant behavioral subtypes and empirical evidence for the computational recognition of ambiverts.