A Research On An Improved ML-KNN Multi-label Classification Method

Multi-label classification is an important branch to research data classification in data mining field.In the era of large data,the surge in the amount of data and the annotation structure of data are becoming more and more complicated,which makes the multi-label learning problem exist in the real world very widely.How to find a fast and effective multi-label classification algorithm with high classification accuracy has become a hot topic in the field of data mining.The research of multi-label data mining is becoming more and more prominent.This paper focuses on the problem of multi-label classification.According to the characteristics of multi-label classification,the main work of this paper is as follows:First,the existing multi-label classification algorithm is summarized and classified.In this paper,the algorithms that have been applied to multi-label classification learning are divided into problem-based transformation strategy and algorithm-based algorithm.For each kind of algorithm,the classification principle,classification step,the advantages and disadvantages of the algorithm and the adaptation conditions are expounded in detail,and several algorithms are simulated on the data set.Second,we propose an improved algorithm:IML-KNN,which is based on the multi-label classification algorithm ML-KNN.The improved points of IML-KNN are described in detail.The classification experiments are carried out on four multi-label data sets,and compared with the other two algorithms.Finally,some factors influencing the algorithm are analyzed and discussed.Finally,A new multi-label classification algorithm is proposed to apply the idea of pseudo-nearest neighbor and penalty function to IML-KNN algorithm.The new algorithm uses the pseudo nearest neighbor(PNN)instead of the nearest neighbor to find the nearest neighbor of the sample x more effectively,and adds the penalty function to improve the posterior probability.Then the principle of pseudo-nearest neighbor and penalty function,the steps of the improved algorithm,the result and analysis of classification experiment are described in detail.