Вычислительные затраты пошаговых методов поиска при идентификации регрессионных моделей
This paper considers the problem of choosing optimal set (subset) of the descriptive variables (regressors) from a fixed set of candidates. Forward Selection and Backward Elimination methods adding/removing a candidate in/from the current set of descriptive variables step-by-step. Each variable is tested to be included or excluded using a chosen model comparison criteria that improves the model the most, and this process repeated until none improves the model. The model selection criteria may be calculated directly or recursively. Algorithms for recursive computing of the residuals sum of squares (RSS) for the model selection criteria in the recursive least squares method are presented.
This paper evaluates the computational costs of the recursive calculation of stepwise model selection criteria for all possible steps of selection.