RE-LABELING METHOD FOR REGRESSION

摘要

According to the present invention, a computer implemented method related to a re-labeling method for a regression analysis comprises: a step (a) of building an initial regression analysis model applying a pre-selected regression analysis method to a predetermined data set wherein the data set is configured as a set of data with a result value and data without a result value; a step (b) of selecting data which is a label target instance in a data set of the built regression analysis model; a step (c) of measuring a result value of the selected data according to the built regression analysis model and, if the selected data is data having no existing result value, providing a new label by a measured result value, and if the selected data has an existing result value, providing a new label by an average value of the existing result value and a newly measured result value; a step (d) of configuring a new data set which updates the data having new labels provided in the existing data set and using the new data set to reconfigure the regression analysis model; and a step (e) of repeating the step (b) to (d) until a predesignated regression analysis termination condition is satisfied, thereby building a final regression analysis model. Accordingly, the present invention may contribute to building a model having a good performance at low costs.