No abstract

No abstract

No abstract

No abstract

During planning and design of new products design managers are interested in estimating cost as early as possible. Unfortunately, in an early design phase only a few attributes of the future product are known, and their impact on cost is not clear to the cost estimation expert. Neural networks might be able to detect hidden relationships between cost drivers and cost of a new product, and estimate cost afer being presented a small set of conceptual attributes describing the product. Based on a laboratory benchmark example with artijcial data we report about our experiments with classlJication perceptrons with one hidden layer. The attention is focused on the problem of small number of training samples available in the domain. The results of three possible remedies are presented, namely prewiring background knowledge, preprocessing input data, and transforming input data.