As a certain type of mixtures of normal densities, a special mixture of regression models has been studied as a switching regression model. Switching regression model is well known and can be found in most texts on multivariate statistics. A family of fuzzy c-regression models (FCRM), recently, is used to study this model. However, FCRM, as a promising technique for switching regression parameter estimation and clustering, still has drawback of slow (or not enough fast) convergence in some cases. In this article, a new algorithm, called Fast FCRM (FFCRM) algorithm is proposed by improving the FCRM for convergence speed. FFCRM, with an additional probability threshold to accelerate convergence, converges much more rapidly than FCRM, and produces high-quality estimates. Two small simulation experiments illustrate the above points.
展开▼
