Multi-label K-Nearest Neighbor Algorithm Based On Granular Computing

With the rapid development of today’s society as well as information technology, multi-label classification learning problem becomes an important classification problem in the real world, which is widely used in text categorization, picture scene classification and classification of gene function. Multi-label classification has become a focus of machine learning, machine learning researchers provide many multi-label algorithms to solve the multi-label classification learning problem. in general, these algorithms could be grouped into two main categories:problem transformation methods algorithm adaptation methods problem. transformation methods, those methods that transform the multi-label classification problem either into some single-label classification or regression problems; algorithm adaptation methods, those methods that transform the multi-label classification problem into some single-label classification problems.This thesis introduces the multi-label classification learning, and then focuses on a number of important multi-label classification algorithms:multi-label classification algorithm based on boosting learning, multi-label classification algorithm based on support vector machines, multi-label classification algorithm based on neural network, multi-label classification algorithm based on covering algorithm and k-nearest neighbor based algorithm for multi-label classification. analysis of the advantages and disadvantages of these algorithms, and propose a new multi-label classification algorithm based on the certain deficiencies of k-nearest neighbor based algorithm for multi-label classification. Mainly work the following:1、The multi-label learning as well as a number of important multi-label classification algorithm on the specific process of these multi-label classification algorithms are given, and point out the advantages and disadvantages of these algorithms, in practical applications and some future improvements of the algorithm has also been discussed2、k nearest neighbors are chosen without considering the distribution of samples in k-nearest neighbor based algorithm for multi-label classification, analyze the shortcoming of the algorithm and propose the improved algorithm.3、the nearest neighbor set is constructed with the controlling of the granular hierarchy, so the nearest neighbors of a sample have high similarity and highly similar samples can be added to nearest neighbor set. The experimental results show that most of the evaluation criteria in new method are better than the tradition multi label learning algorithm.