Research On The Construction Method Of Hierarchical Monotonic Decision Tree

The decision tree algorithm is an effective machine learning algorithm that performs well in classification tasks.Monotonic decision trees are a special type of decision tree algorithm.Unlike traditional decision tree algorithms,in monotonic decision trees,each non-leaf node represents a feature and the decision outcome changes monotonically with the feature’s value.Many categories in monotonic data have a hierarchical structure.In the hierarchical structure of categories,the closer to the root node,the higher the classification accuracy,but the category range is more general.The closer to the leaf node,the more specific the category range and the more difficult to predict,and the lower the classification accuracy.Addressing the issue where traditional decision tree algorithms only consider accuracy and not precision,the main work of this paper includes the following two aspects:(1)To construct a decision tree from data with monotonicity and hierarchical structure,it is necessary to consider how to classify the data when partitioning attributes,so that the classification results are as specific as possible and the prediction accuracy is high.A hierarchical monotonic decision tree method is proposed for learning classifiers from monotonic data with hierarchical class structure.Experimental results demonstrate the effectiveness of the proposed method in terms of prediction accuracy and precision.(2)To further improve the generalization ability of hierarchical monotonic decision tree models,research is conducted on how to build ensemble hierarchical monotonic decision tree models.To do so,a monotonic attribute reduction method is introduced and a strong learner is obtained through ensemble strategies.Firstly,a rank-preserving attribute reduction method based on variable dominance rough sets is used to generate monotonic attribute reductions and learn base classifiers based on hierarchical monotonic decision trees.Then,according to the principle of combining base classifiers with maximum probability,the ensemble learning system’s generalization ability is further improved.Experimental results show that the ensemble hierarchical monotonic decision tree can improve the classification performance.In summary,to handle monotone data with category hierarchy structure,this paper proposes a hierarchical monotone decision tree method and improves the performance of the decision Tree through integration strategy.Finally,experiments are conducted on three publicly available datasets,and the results demonstrate the effectiveness of the proposed method.