Abstract
<title>Abstract</title> <p>Automated personality assessment has emerged as a critical research frontier at the intersection of computational psychology and machine learning, with applications ranging from human resources and adaptive systems to mental health screening and forensic analysis. This paper presents a robust machine learning pipeline designed to classify individuals as introvert or extrovert using engineered behavioral and social interaction metrics. We draw upon a dataset comprising 2,900 instances, each characterized by eight behavioral attributes, including social engagement frequency, recovery patterns, and digital activity indicators. Our methodological contribution lies in the systematic en-gineering of three composite features—social ratio, social engagement, and recovery need—designed to capture latent psychological constructs not directly observable in raw questionnaire responses. Through comprehensive ablation studies, we quantitatively demonstrate the contribution of these engineered features, showing that they provide a 12.4% improvement in classification accuracy com-pared to using only original features. Our experimental framework rigorously evaluates 13 distinct learning algorithms, ranging from linear and ensemble models to neural networks and probabilistic classifiers. The best performing model, K-Nearest Neighbors, attains an accuracy of 91.90%, while RBF Support Vector Machine and Gaussian Naive Bayes follow closely at 91.72%. We perform in-depth analysis of why K-NN performs best, attributing its success to the well-clustered nature of the engineered feature space. Paired t-tests confirm that differences among top models are not statistically significant (p > 0.05), providing practitioners flexibility in model selection. Feature importance analysis ranks the engineered features among the most influential predictors, validating our feature engineering hypothesis. This study contributes validated, theoretically-grounded models to computational psychology, with immediate practical implications and clear pathways for future research.</p>