Sparse incremental regression modeling using correlation criterion with boosting search

A novel technique is presented to construct sparse generalized Gaussian kernel regression models. The proposed method appends regressors in an incremental modeling by tuning the mean vector and diagonal covariance matrix of an individual Gaussian regressor to best fit the training data, based on a correlation criterion. It is shown that this is identical to incrementally minimizing the modeling mean square error (MSE). The optimization at each regression stage is carried out with a simple search algorithm re-enforced by boosting. Experimental results obtained using this technique demonstrate that it offers a viable alternative to the existing state-of-the-art kernel modeling methods for constructing parsimonious models.