Interpretability of automated machine learning methods in psychological research: A tutorial with AutoGluon in Python.

Integrating artificial intelligence into psychological research represents a significant direction in contemporary psychology. Utilizing supervised and unsupervised machine learning techniques can further aid in understanding the nonlinear relationships of psychological concepts. In machine learning, variables, referred to as features, can encompass data from psychological scales, text, audio, and images. Current psychological research predominantly relies on frequentist approaches, where relationships between variables are typically based on regression, which often falls short in handling the nonlinear relationships of psychological characteristics. Therefore, we outline an innovative semi-automated workflow that empowers psychology researchers to leverage machine learning algorithms for intelligent model selection, facilitating the construction of more precise and insightful theoretical frameworks. This approach aims to achieve three primary research objectives: (1) automated hyperparameter tuning to attain optimal models; (2) identification of important features through interpretability techniques, facilitating feature selection based on calculated importance; (3) data-driven insights for theory building based on important features by integrating exploratory factor analysis with machine learning interpretability. In this paper, we provide an introduction to the basics of machine learning, describe the benefits of combining automated machine learning for researchers, and, using psychological resilience research as an example, offer a detailed annotated code workflow along with raw data. This low-code approach, designed with psychological research methodologies in mind, makes it highly accessible for psychological researchers.