Fast Multi-output Regression Based on Piecewise Linear Approximation

In this paper a fast Multi-output Regression algorithm is implemented based on piecewise linear approximation (PLA). Maximum distance point splitting is used to fit lines piecewisely. Then the best knot set of linear segmentation is obtained. The independent variable X is divided into some linear intervals, which make it possible that linear regression is implemented in fast and accurate way. The knot set with the smallest error in all dimensions of dependent variable is used as the node partition scheme. The matrix of regression model is derived by the least square. The Experiment results prove that the algorithm achieves the performance of the state-of-the-art with predicting time below millisecond.