Machine Learning–Driven Adaptive Testing: An Application for the MMPI Assessment

Abstract

This paper aims to examine the effectiveness of machine learning classification algorithms as a strategy to overcome the limitations associated with traditional methods for developing computerized adaptive versions of the Minnesota Multiphasic Personality Inventory-2 (MMPI-2). The focus is on the three scales in the neurotic area of the instrument, namely, hypochondria, depression, and hysteria, which were administered electronically to a nonclinical sample of 383 participants. The findings indicate that a machine learning classifier based on a model tree (ML-MT) algorithm effectively handled the complex MMPI-2 scales, yielding accurate scores while noticeably reducing item administration. In particular, the ML-MT algorithm achieved item savings between 85.99% and 93.78% and produced scores that differed from those of the full-length scales by only 2.5–3.3 points. Compared to the countdown algorithm, the ML-MT algorithm proved to be significantly more efficient and accurate. Furthermore, the ML-MT scores retained their validity, as indicated by correlations with other MMPI-2 scales that were comparable to those obtained with the full-length scales (the average difference between the correlations was less than 0.10). These findings support the potential of the ML-MT algorithm as an effective method for adaptive assessment in the context of the MMPI instruments and other psychometric tools.