Method And Application For Multi-target Regression Via Target Specific Features

The traditional regression method assumes that a sample has only one regression target.However,in a large number of regression scenarios,a sample usually has multiple transaction information,corresponding to multiple regression targets.The traditional single-target regression methods lack the capabilities to accurately and effectively express the complex information of the sample.Multi-target regression(MTR)is a regression method for data samples with multiple regression targets at the same time.Therefore,multi-objective return to nature has become an important means and method to solve the problem of multi-information data regression.In MTR,a sample belongs to multiple targets at the same time,and it can better represent the diversity of information.In recent years,multi-objective regression has received great attention in many fields such as computer vision and medical image analysis,and it has considerably potential applied value in the real world.Although the development of MTR is changing with each passing day,existing MTR methods are based on the same input(feature)space to predict all targets,but different targets may have different characteristics on their own unique characteristics.Thus,these methods may be only suboptimal.In this paper,we propose a method that employs target-specific features to predict different regression targets.At the same time,this method is applied to two important challenges in the field of MTR: 1)how to model the correlation among targets;2)how to deal with complex input-output relationship.In view of the above problems,the main research contents of this paper can be divided into the following three aspects:1.MTR datasets have multiple regression targets at the same time.In order to effectively model the correlation among targets,improving the experimental performance.Firstly,a hierarchical clustering method is applied and samples with similar characteristics are assigned to the same leaf nodes,achieving the purpose of modeling correlation among targets.2.In order to effectively deal with complex input-output relationship,we propose a MTR method via target-specific features.The method uses a classification regression tree to calculate a similarity dependence matrix for each target,and then performs a clustering operation through the corresponding similarity dependency matrix to extract appropriate and distinctive features for each target.3.In today's society where information technology is highly developed,e-commerce electronically and digitally transforms traditional business processes.On the one hand,electronic logistics replaces real logistics,which can greatly reduce expenditures on manpower and material resources,reduce costs,and improve efficiency.At the same time,a large amount of data is generated,and the rational use of data information and the value existing inside the information can improve the knowledge and understanding of things,further optimize resource allocation,and improve economic and social benefits.In order to apply multi-target regression to real-world engineering,this paper applies multi-target regression based on target-specific features to the competition dataset from Alibaba IJCAI17 to verify the stability of the model.The method improves the performance of the algorithm by exploiting the correlation between the pertinent and discriminative features of each target.In order to verify the validity of the MTR method proposed in this paper,we perform experiments on 18 datasets of which results show that the method achieves competitive performance against representative state-of-the-art MTR methods.At the same time,in order to verify the effect of the method in practical applications,we conducted experiments on a realistic engineering application from Alibaba,and receive excellent performance.