Abstract
As a new application of GMDH, a largespatial pattern identification problem is solved where space is the independent variable of the system. The interest to solve this problem has been stimulated by the practical situation of monitoring air pollution concentration. Selection of input variables is one of the most important tasks in the GMDH. Two kinds of input variables are tested for our purpose:
Type 1: (x, y) coordinates
Type 2: (x, y) coordinates and the output of pointwise information at the neighboring points of a prediction point.
Although Type 2 involves more computational tasks than Type 1, we can obtain better accuracy of pattern identification from Type 2.
GMDH algorithm is implemented in the Time Sharing System (TSS), therefore we could use the computer program by the man-computer interactive way which enables us to change some heuristics in the GMDH easily.
