Slant Split Criterion Random Forests Classification Algorithm based on Soft Margin Hyperplane
Xueda Kan, Tiansong Li · 2020
In order to further to improve the classification accuracy of the random forest algorithm, an Slant Split Criterion Random Forests based on Soft Margin Hyperplane (SSCRF) algorithm is proposed. In the decision tree node splitting process, consider using multiple attribute soft margin hyperplane of the decision tree to replace a single attribute splitting criterion, so that the splitting criterion becomes an inclined hyperplane in the data space, which effectively improves the accuracy of decision tree classification; In order to make the algorithm suitable for multi-classification problems, the "one-to-many" strategy is adopted in the process of node splitting; in the process of transforming into leaf nodes, various adaptive weights are assigned to improve the algorithm's classification ability. The experimental results show that the SSCRF algorithm has higher classification accuracy than the random forest algorithm.