Adaptive Fit Parameters Tuning with Data Density Changes in Locally Weighted Learning

Abstract

Locally weighted learning (LWL) is a form of lazy learning and focuses on locally weighted regression. Due to its high efficiency and flexibility, the learning mechanism is widely used in prediction. However, LWL fails when the data points are sparse, and fewer survey concerns about tuning fit parameters in local model with density of the data input. This paper discusses the relationship between data density and fit parameters from a theoretical view. The relationship we advocate also contributes to adaptive fit parameters selection. Experimental studies provide evidence for the mathematical derivation and show its application superiority in prediction of traffic flow.