Abstract
In this paper, we propose the ensemble prediction method by using Self-Organizing-Maps (SOM) with the prediction reliability measurements. Splitting the training data to several groups, creating maps, and testing about each other groups obtained the prediction reliability measurements. Another group tested the map of certain group, and then the values (codebook vectors) of the tested nodes were evaluated. Collecting the nodes value and the reliability measurements about tested nodes, and inputting the values as new training data to SOM, then the map with prediction reliability measurements was obtained. Furthermore, difference models were able to create by setting various priority factors (weight) of the variables. The ensemble prediction was carried out by prediction values of these models.
