Revised GMDH Algorithm Using Prediction Sum of Squares (PSS) as a Criterion for Model Selection
- 1 January 1978
- journal article
- Published by The Society of Instrument and Control Engineers in Transactions of the Society of Instrument and Control Engineers
- Vol. 14 (5) , 519-524
- https://doi.org/10.9746/sicetr1965.14.519
Abstract
In this paper, a revised GMDH (Group Method of Data Handling) algorithm is developed in which we do not require to divide available data into two groups; training data and checking data. In this algorithm, all the data can be used not only as the training data but as the checking data, that is, the Prediction Sum of Squares (PSS) calculated from all the data is used as a criterion for selecting intermediate variables and for stopping the multilayered calculation. Therefore, the identified results do not depend on the heuristics of dividing the data into two groups. Furthermore, the revised GMDH developed in this paper automatically generates optimal partial polynomials in each selection layer. The revised GMDH, therefore, has flexibility much better than that of the basic GMDH in constructing a complete polynomial. The revised GMDH algorithm is applied to a simple illustrative example and compared with the result obtained by the basic GMDH algorithm.Keywords
This publication has 0 references indexed in Scilit: