An explainable artificial intelligence handbook for psychologists: Methods, opportunities, and challenges.

Area of Science:

• Psychology
• Computer Science
• Artificial Intelligence

Background:

• Machine learning is increasingly used in psychology for large datasets.
• Understanding model predictions and key predictors requires explainable artificial intelligence (XAI).
• Methodological approaches for predictor importance in machine learning are less established than in traditional statistics.

Purpose of the Study:

• To introduce explainable artificial intelligence (XAI) to psychologists.
• To provide an overview of common XAI methods.
• To analyze the impact of multicollinearity on XAI methods in psychological research.

Main Methods:

• Overview of applied machine learning explainability vs. psychological explainability.
• Review of XAI techniques: permutation importance, impurity-based feature importance, individual conditional expectation (ICE) graphs, partial dependence plots (PDP), accumulated local effects (ALE) graphs, Local Interpretable Model-agnostic Explanations (LIME), SHapley Additive exPlanations (SHAP), and Deep Learning Important FeaTures (DeepLIFT).
• Simulation analysis to demonstrate multicollinearity's impact on XAI methods.

Main Results:

• Multicollinearity can significantly impact the results of various XAI methods.
• Different XAI methods exhibit varying sensitivities to multicollinearity.
• Challenges in implementing XAI for psychological data were identified.

Conclusions:

• XAI offers valuable tools for understanding machine learning models in psychology.
• Careful consideration of method choice and data characteristics (e.g., multicollinearity) is crucial for reliable predictor importance.
• Further research is needed to address implementation challenges and advance XAI applications in psychology.