Multi-label Classification Model:Combining Instance-based Model And Logistic Regres-sion

In the era of’big data’, how to effectively use these data become an urgent prob-lem to solve. Classification is one of the common direction to use data. Especially, Multi-label classification is an extension of conventional classification in which a single instance can be associated with multiple labels. Multilabel classification has received increasing attention in recent years, due to its practical relevance. Recently, quite a number of algorithms for multi-label classification has been proposed based on existing classification algorithms. However, existing algorithms do not take correlations and interdependencies between labels into account, and the performance of existing models can be improved.In this paper, we propose a new approach to multi-label classification, which combines instance-based learning and logistic regression, using the information coming from the neighbors of the instance as a additional feature, we combine it with the original features, then build the special logistic regression model. This approach allows one to capture interdependencies between labels and combine the advantages of both methods. In the experimental studies, first we give a summary of several evaluation criteria for multi-label prediction, then we compare our approach with some existing algorithms of multi-label classification using the data from an e-commerce company.