Abstract
The emergence of big data greatly promotes the development of data-driven machine learning technologies. The common assumption in machine learning that the training data and the test data have identical feature spaces with underlying distributions impedes the development of machine learning. To deal with this issue, transfer learning is studied to exploit the knowledge accumulated from data in auxiliary domains to facilitate predictive modelling consisting of different data patterns in the current domain. There have been a significant number of methods proposed to address the classification, as the task, through transfer learning, but the studies targeted at regression problems are still scarce. In this paper, we propose a new transfer learning method to deal with the regression task in the target domain where few data are available. Takagi-Sugeno fuzzy model is used to construct the model for regression task in the source domain, and the prototypes and linear functions of the existing model are modified to make the model more compatible for the target domain. The experimental results demonstrate that our method can improve the performance of the existing model of the source domain on addressing current task in the target domain.