Feature selection for semi-supervised multi-target regression using genetic algorithm

Multi-target regression (MTR) is an exciting area of machine learning where the challenge is to predict the values of more than one target variables which can take on continuous values. These variables may or may not be correlated. Such problems commonly occur in real life scenarios, and therefore, interest and research in this area has increased in recent times. Some examples of applications include analyzing brain-activity data gathered using multimedia sensors, stock information from continuous web data, data related to characteristics of the vegetation at a certain site, etc. For a real-world multi-target learning system, the problem can be further complicated when new issues emerge with very little data available. In such cases, a semi-supervised approach can be adopted. This paper proposes a Genetic Algorithm (GA) based semi-supervised technique on multi-target regression problems to predict new targets, using very small number of labelled examples by incorporating GA with MTR-SAFER. Experiments are carried out on real world MTR data sets. The proposed method isexplored with different variations and also compared with the state of the art MTR methods. Results have indicated a significantly better performance with the further benefit of having a reduced feature set.