A Review of Interpretability Methods for Gradient Boosting Decision Trees

Download PDF
Open Webpage

Victoria Sousa Figueira Gonçalves, Vinicius Renan de Carvalho

Published: 2025, Last Modified: 31 Oct 2025J. Braz. Comput. Soc. 2025EveryoneRevisionsBibTeXCC BY-SA 4.0
Abstract: This survey examines interpretability methods used or proposed for Gradient Boosting Decision Trees, which are advanced machine learning algorithms based on decision trees. The studies analyzed were gathered using synonyms for "explainability" combined with synonyms for "method," as well as synonyms for "Gradient Boosting Decision Trees." The proposed or applied approaches are classified by their techniques and described in detail. Among these methods, we recommend using SHAP values to rank features based on their relevance, as this approach aligns well with the structure of Gradient Boosting Decision Trees. Additionally, we suggest considering inTrees, RULECOSI+, and Tree Space Prototypes when applicable.