A Multivariate Decision Tree Classification Algorithm With Quantitative Monotonicity Constraints

Decision tree is a kind of data mining method of case classification in an intelligent way,which has been widely used in artificial intelligence and other fields.Among them,the monotone classification algorithm can solve the classification problem of monotone order relation between attribute and label value.Monotonic order relation refers to a monotonic constraint on attributes and labels in the process of classification,that is,the attribute value increases,and the class label value should remain unchanged or increase correspondingly.This kind of classification problem widely exists in the field of customer credit rating evaluation.However,previous experiments show that the traditional monotone classification algorithm is usually sensitive to noise data and has some limitations on the types of data sets.In order to solve the above problems,this paper proposes a decision tree classification algorithm based on the distribution of data sets.Firstly,the original data set is denoised according to the index of non similar quantity ratio to reduce the impact of noise samples on the classification results,so as to improve the classification accuracy.Secondly,because the traditional monotonic classification method is only applicable to the data set with ordered attributes and labels,and the single variable decision tree can not comprehensively consider the impact rate of all attributes on the classification task,so this paper maps the data set into multi-dimensional space,and combines the method of local density peak clustering to form a monotonic classification with quantitative monotonic constraints,thus evolving into a multivariate decision tree,which can not only eliminate the constraints on the types of data sets,but also incorporate the influence of all attributes into the classification process.Finally,the optimal split hyperplane is determined by the relative boundary points and the sum of local misclassification rate.In the aspect of experiment,this paper applies the multi variable decision tree classification algorithm with quantitative monotone constraints to 11 data sets selected from UCI.At the same time,the proposed algorithm is compared with ID3 algorithm(REPtree),C4.5 algorithm(J48),RandomForest,RandomTree and HoeffdingTree.The experimental results show that the proposed classification algorithm is superior to other decision tree classification algorithms in terms of classification accuracy,mean absolute error and F1-Measure,so it has good classification performance.