Improved linear regression prediction by transfer learning

David Obst, Badih Ghattas, Sandra Claudel, Jairo Cugliari, Yannig Goude, Georges Oppenheim

Abstract: Highlights•A new approach of parametric transfer is proposed for the linear model.•The benefits of transfer are quantified using the gain in prediction error.•A statistical test is proposed to predict in advance whether transfer is beneficial or not for a new sample x.•The methodology has been applied on synthetic and real-world data with success.