Fuzzy Factor in Switching Regression Models

Switching regression models is a kind of special mixture regression models, and the clustering method of it can be classified as hard partitioning and fuzzy partitioning. The former is simple in coding and fast in execution, but its partitioning is too pure since it overlooks the fuzziness; whereas the latter, taking the fuzziness into account, is more reasonable for partitioning, but the algorithm complexity and cost will increase. Based on the research of hard partitioning and fuzzy partitioning, the idea of fuzzy factor is proposed. By adjusting fuzzy factor, hard partitioning and fuzzy partitioning are unified, and the clustering analysis of switching regression models becomes more flexible. It is demonstrated by simulation experiment that it is feasible to find the optimal combination of speed and fuzziness according to the practical application.