A case study on target cost estimation using a genetic algorithm and a back-propagation based neural network

Establishing the target cost of new products has always been difficult, as only a few attributes of the product as usually known. In these circumstances, parametric methods are commonly used by using a predetermined cost function where the considered parameters are evaluated from historical data. In contrast to the regression or parametric models, neural networks, in are non-parametric which attempt to fit curves to predict the cost without being provided a predetermined function. In this paper, the above mentioned property of neural networks is used to investigate their applicability for cost estimation of a certain major aircraft component. This empirical study is conducted in collaboration with a major aerospace company located in Montreal, Canada. Two neural network models, one trained by the gradient decent algorithm and the other by genetic algorithm, are considered and contrasted to one another. The study, using historical data, shows that the neural network model trained by genetic algorithm outperforms the model trained by back-propagation as it fits well both in the training and validation data sets.