Multi-label Classification Algorithm Based On Random Forest And Predictive Clustering Tree

In recent years,multi-label classification problems become more and more common in life,and also gradually become a hot spot in academic research.In multi-label classification problem,each sample in the training set corresponds to one or more label combination,while in traditional classification problems,each sample belongs to only one category,they are significantly different from multi-label classification problems.Therefore,the algorithm for single-label classification cannot be directly used for multi-label classification problem.At present,many algorithms have been used to solve multi-label classification problems,but after our research and experiment,we find that they do not reach the expected results,there are still many room for improvement.Based on the research of network regression,this paper applies the method of network regression to the predictive clustering tree algorithm,and then uses the random forest algorithm to integrate it,thus proposes a new multi-class standard classification algorithm.The main work of this paper includes:(1)The domestic and international literatures are studied from three aspects: random forest,forecasting clustering tree and multi-class classification algorithm.The classification algorithm and its verification criterion in multi-class standard research field are summarized comprehensively.(2)By transforming the non-network data into the network data,the algorithm used in the network regression is applied to multi-class classification scenarios.(3)Combining the distance between clusters and the internal tightness calculated by the method of network regression as the basis for predicting the optimal splitting property of the clustering tree;(4)Based on the above work,a new multi-class classification algorithm is proposed by integrating the predictive clustering tree algorithm with random forest algorithm.In the end,the newly proposed method is compared with many excellent algorithms which are commonly used in the industry on several data sets for multiple domains.The experimental results show that our algorithm has good classification performance.