Automated machine learning for classification and regression: A tutorial for psychologists.

Area of Science:

• Psychology
• Computer Science
• Data Science

Background:

• Machine learning (ML) offers data-driven insights in psychology but faces adoption barriers due to complexity and lack of standardization.
• The 'black-box' nature of ML hinders understanding variable influence, limiting its application in psychological research.
• Automated ML (AutoML) and explainable AI (XAI) can address these challenges by automating processes and improving transparency.

Purpose of the Study:

• To introduce Automated ML (AutoML) and explainable AI (XAI) to psychologists, bridging the gap in educational resources.
• To demonstrate the practical application of AutoML in psychological research using the "H2O AutoML" R package.
• To provide guidance on advanced AutoML methods and workarounds for unsupported ML techniques.

Main Methods:

• The study covers advanced AutoML techniques, including combined algorithm selection and hyperparameter optimization (CASH) and stacked ensemble generalization.
• Utilized the "H2O AutoML" R package for practical demonstrations with psychological datasets.
• Applied regression on multi-individual cross-sectional data and classification on single-individual time-series data.

Main Results:

• AutoML streamlines ML workflows, making advanced methods accessible to researchers with varying technical expertise.
• Demonstrated successful application of AutoML for both regression and classification tasks in psychological data analysis.
• Provided practical solutions for implementing ML methods not directly supported by the "H2O AutoML" package.

Conclusions:

• AutoML democratizes advanced machine learning for psychological research, enhancing data analysis capabilities.
• AutoML and XAI empower psychologists to leverage complex datasets for improved prediction and discovery.
• This work facilitates the adoption of powerful ML tools, advancing the field of psychological science.