Constrained center and range joint model for interval-valued symbolic data regression

A constrained center and range joint model to fit linear regression to interval-valued symbolic data is introduced. This new method applies both the center and range of the interval to fit a linear regression model, and avoids the negative value of the range of the predicted dependent interval variable by adding nonnegative constraints. To improve prediction accuracy it adopts overlapping constraints. Using a little algebra, it is constructed as a special case of the least squares with inequality (LSI) problem and is solved with a Matlab routine. The assessment of the proposed prediction method is based on an estimation of the average root mean square error and accuracy rate. In the framework of a Monte Carlo experiment, different data set configurations take into account the rich or lack of error, as well as the slope with respect to the dependent and independent variables. A statistical t-test compares the performance of the new model with that of four previously reported methods. Based on experiment results, it is outlined that the new model has better fitness. An analysis of outliers is performed to determine the effects of outliers on our proposal. The proposed method is illustrated by analyses of data from two real-life case studies to compare its performance with those of the other methods.