Smooth approximation algorithm based on split-merge model trees
Wei Wang · Journal of Tsinghua University(Science and Technology) · 2003
A smooth approximation alg or ithm based on split-merge model trees was developed for sample data-based mode ling of nonlinear functions. The local linear model trees algorithm was used to partitions the input space into several areas using the split-merge algorithm. A piece-wise linear function was used to approximate and construct weighting f unctions in each region. The full expression of the basis functions was then use d to obtain smooth approximations with arbitrary precision. The split-merge alg orithm simultaneously applies some of the linear functions on several non-conve x or non-connected regions. A comparison using the same number of parameters as in the local linear model trees algorithm shows that this smooth approximation algorithm enhances the approximation precision.