A Mixture of Experts with Two-Stage Adaptive Fuzzy c-means Gate
Jian Jin, Long Jin · 2011
A mixture of experts with a two-stage adaptive FCMef (2-AFCMef) gate is presented to overcome problems of equal-sized divide and conquer and its efficiency. In 2-AFCMef algorithm, a descending ordered series of indexes of effectiveness factor is applied in turn for trying, so that the maximal index can be achieved while any cluster does not disappear. The result of traditional fuzzy c-means clustering algorithm, instead of blindly and randomly searching for many times, is pre-treated as the initial condition of FCMef algorithm. The new method is applied to the field typhoon tracks prediction in South China Sea. Results show great improvement by more than 10% compared to that of the traditional method under the same conditions.