Research On Dimensionality Reduction Algorithm Based On K-nearest Neighbor Multi-label Learning

With the development of "Internet plus",multi-label data is widely produced,which is closely related to people's life applications.Due to it needs to consider more features and the correlation between labels and labels in the process of learning.Therefore,it is more complex and challenging than single label learning to data classification and data dimension.Since the last 1990 s,the concept of the multi-label learning has been proposed.It has attracted the attention of many experts and scholars for the research and the results have mushroomed.Multi-label learning mainly concentrated on two research direction for the classification and dimension reduction.Data dimension reduction is an important step in machine learning,and it is an important means to improve the performance of data classification.In this paper,PCAI and MRF-mRMR algorithms are propose for data dimension reduction.Compared with the original algorithm,the classification effect of the PCAI algorithm is improved significantly.The MRF-mRMR algorithm can remove redundant feature features at the same time also keep the property and the correlation between labels.In this paper,the study of multi-label learning can be divided into two parts.The first part puts forward the PCAI data dimension reduction algorithm based on PCA algorithm,and uses ML-k NN classifier for data classification.Firstly,we put forward the conception of tolerance,and the computation formula is defined.At the same time we also has carried on the experiment to investigate the parameters in the formula.Secondly,we get the eigenvalues of the data dimension reduction,and use characteristic value to ML-kNN classifier for weighted distance.Finally,after drop for the dimensional data set is applied to the improved ML-kNN classifier to verify the dimension reduction effect.The second part for Relief and mRMR algorithms are described in detail.On the one hand,characteristics of Relief algorithm for feature weight calculation method was improved,on the other hand,MRF-mRMR algorithm combined with feature selection is proposed.MRF-mRMR algorithm that we proposed is not only maintains the advantage of maximal correlation and the minimum redundancy,but also owns the advantages of Relief algorithm to sort each feature weight.The experimental results show that the MRF-mRMR algorithm's effect is better than existing algorithms for dimension reduction.To sum up,this paper presents two kinds of dimension reduction algorithms,PCAI and MRF-mMMR,and takes advantage of ML-kNN classifier for dimension reduction effect.The experimental results show that the data dimension reduction effect is obvious,and data classification effect is improved.