Binary Split Categorical feature with Mean Absolute Error Criteria in CART

The recent publication on arXiv introduces a groundbreaking approach to the Classification and Regression Trees (CART) algorithm, focusing on the splitting of categorical features using the Mean Absolute Error (MAE) criterion. Traditionally, the use of MAE has been hampered by the reliance on various numerical encoding methods, which this paper argues are not suitable for categorical data. By presenting a novel and efficient splitting algorithm, the authors address these challenges, emphasizing the limitations of existing methods. This advancement is crucial for enhancing the handling of categorical data in machine learning, potentially leading to more accurate and effective models.