Multivariate Regression Forest for Categorical Attribute Data

• Published in: Ji suan ji ke xue Vol. 49; no. 1; pp. 108 - 114
• Main Authors: Liu, Zhen-yu, Song, Xiao-ying
• Format: Journal Article
• Language: Chinese
• Published: Chongqing Guojia Kexue Jishu Bu 01.01.2022; Editorial office of Computer Science
• Subjects: Classification; Clustering; Decision analysis; Decision trees; decision trees|multi-variable regression trees|ensemble learning|random forest|gradient boosting; Multivariate analysis; Regression analysis; Regression models; Root-mean-square errors
• ISSN: 1002-137X
• DOI: 10.11896/jsjkx.201200189

Aiming at the problem that regression models such as linear regression, SVR and most multivariate regression trees cannot directly use classification attributes for regression analysis, a decision tree node division method that can combine multiple types of attributes is proposed. The center of the set on the classification attribute and the distance from the sample to the center, so that the classification attribute can also participate in the clustering process of the sample like the numerical attribute, so as to form the division of the sample set. The decision tree selects an appropriate ensemble scheme, and the resulting ensemble is called Cluster Regression Forest (CRF). Finally, the regression mean absolute error (MAE) of CRF and other 9 regression models is compared on 12 UCI public datasets and root mean square error (RMSE), the experimental results show that CRF has the best performance among the 10 regression models.