Frequency-adjusted borders ordinal forest: A novel tree ensemble method for ordinal prediction.

Area of Science:

• Psychological sciences
• Machine learning applications in psychology
• Statistical modeling for ordinal data

Background:

• Ordinal responses are prevalent in psychology (e.g., rating scales, grades).
• Traditional methods like proportional odds models are being supplemented by machine learning (ML).
• Random Forest (RF) is a popular ML method, but requires adaptation for ordinal data.

Purpose of the Study:

• To introduce a novel tree ensemble method, Frequency-Adjusted Borders Ordinal Forest (fabOF), for ordinal prediction.
• To improve predictive performance in ordinal data analysis without computationally intensive optimization.
• To provide an interpretable variable importance measure tailored for ordinal prediction.

Main Methods:

• Proposed Frequency-Adjusted Borders Ordinal Forest (fabOF), a novel tree ensemble method.
• fabOF bypasses extensive optimization procedures common in other RF extensions for ordinal data.
• Introduced a permutation variable importance measure specifically designed for fabOF and ordinal outcomes.

Main Results:

• fabOF demonstrated improved predictive performance in simulations compared to existing methods.
• An illustrative example of student performance prediction showed the efficacy of fabOF.
• Key predictors for student performance identified: interest in higher education, mother's education, and study time.

Conclusions:

• fabOF offers a computationally efficient and effective alternative for ordinal prediction in psychological research.
• The method enhances predictive accuracy while providing interpretable insights into predictor importance.
• The fabOF methodology is available via an accompanying R package for broader adoption.