Multi-view Adaptive K-nearest Neighbor Classification Based On Decision Tree

With the rapid development of information technology and related fields,the ways of acquiring data have become increasingly diversified.The same data information can be obtained from multiple perspectives,and the data obtained in this way is called multi-view data.Compared to single-view data,multi-view data contains richer sample feature information.In the classification of multi-view data,the data between multiple views can be consistent and different.Traditional single-view machine learning algorithms are no longer able to meet the requirements of multi-view classification,and more and more relevant improved multi-view algorithms are being proposed.The Adaptive K-Nearest Neighbor(AKNN)algorithm can provide different k values for samples,and its classification rules are simple and interpretable,making it suitable for a variety of data classification tasks.For multi-view classification tasks,this paper proposes two algorithms: Multi-view Adaptive K-Nearest Neighbor(MVAKNN)based on decision tree and Random Forest based Multi-view Adaptive KNN(RF-MVAKNN).The AKNN algorithm obtains the optimal k value for each sample through a sample correlation measurement matrix.In consideration of the influence of sample labels on the distance between samples,the inter-class and intra-class distances are introduced into the correlation matrix.The adaptive KNN algorithm based on decision tree is extended to the multi-view classification field,and the Dempster-Shafer evidential theory combination rule is used to effectively combine the information of multiple views,improving the algorithm’s classification accuracy under multi-view data.To address the issue of constructing a single decision tree in the MVAKNN algorithm,the RF-MVAKNN algorithm utilizes a random forest to replace the single decision tree,which enhances the robustness of the proposed algorithm.The main innovations of this paper are as follows:(1)The sample labels are added to the process of calculating the sample correlation weight matrix,and the intra-class and inter-class distances are introduced as criteria for judging sample correlation,so that samples with the same class label have smaller weights.(2)The adaptive KNN algorithm based on decision tree is extended to the multi-view classification field,improving the multi-view classification accuracy.(3)The Dempster-Shafer evidential theory is used to fuse and calculate the output of each view,which not only combines the classification information of multiple views but also gives the credibility of the classification results.(4)By using random forests,the impact of minor data changes on decision tree construction is eliminated,improving the robustness of the algorithm.In addition,a threshold is added when calculating the correlation matrix to optimize the calculation of k value in the training part and further improve the multi-view classification accuracy of the algorithm.In this paper,the adaptive KNN algorithm based on decision tree for multi-view data classification is studied,and the classification effect of the adaptive KNN algorithm is optimized.The problem of single-view AKNN classification algorithm’s inability to combine multiple view information in multi-view classification is also solved.