Psychometric properties of personality assessment using machine learning

Technological advancements have enabled personality psychologists to move beyond traditional questionnaire-based assessment toward machine learning-based personality assessment (ML-PA). This manuscript provides a non-systematic overview of the validity and reliability of ML-PA, where behavioral features (e.g., text, voice, digital footprints) serve as predictors of personality traits. ML-PA shows promising construct validity, particularly for observer reports, and ML-PA values are similarly correlated with external variables as questionnaire-based values. However, reliability indices, especially for self-report-based ML-PAs, have been found to be lower. Factors such as sample size, input data quantity, and trait activation significantly impact ML-PA accuracy. Algorithmic bias might pose a threat to ML-PA, and there is a trade-off between applying bias mitigation techniques and maximizing ML-PA performance. Future advancements, including the use of large language models and a focus on explainability, are expected to further enhance personality measurement using computational methods.