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

Area of Science:

• Psychology
• Computer Science
• Data Science

Background:

• Contemporary psychology increasingly integrates artificial intelligence (AI).
• Traditional frequentist approaches struggle with nonlinear relationships in psychological data.
• Machine learning (ML) offers advanced methods for analyzing complex psychological variables.

Purpose of the Study:

• To present a semi-automated workflow for psychology researchers to utilize ML.
• To enable intelligent model selection for more precise theoretical frameworks.
• To address limitations of frequentist methods in capturing psychological nuances.

Main Methods:

• Implementation of a semi-automated workflow for ML in psychological research.
• Automated hyperparameter tuning for optimal model performance.
• Feature selection using ML interpretability techniques and exploratory factor analysis.
• Integration of ML interpretability with exploratory factor analysis for data-driven theory building.

Main Results:

• The workflow facilitates automated model optimization and feature identification.
• Interpretability techniques reveal important psychological features.
• Exploratory factor analysis combined with ML insights aids theory construction.
• A low-code approach makes ML accessible to psychology researchers.

Conclusions:

• The proposed workflow enhances psychological research by leveraging AI and ML.
• This approach improves the precision and insightfulness of theoretical frameworks.
• It democratizes the use of advanced computational methods in psychology.
• The methodology is demonstrated with an example in psychological resilience research.