Applying Neural Networks in Case-Based Reasoning Adaptation for Cost Assessment of Steel Buildings

Abstract Case-based reasoning (CBR) refers to both a cognitive and a computational model of reasoning by analogy from past cases. It is often more efficient to solve a problem by starting with the solution(s) to a previous similar problem than it is to generate the entire solution again from first principles. In fact, experts have been observed to reason by analogy to prior cases. Case adaptation, a central component of case-based reasoning, is often considered the most difficult part of a case-based reasoning system. The difficulties arise from the fact that adaptation does not often converge, especially if it is not done in a systematic way. This problem, sometimes termed the assimilation problem, is especially pronounced in the case-based design problem-solving domain where a large set of constraints and features are processed. Furthermore, in the steel building design domain, multiple cases must be considered in conjunction in order to solve the new problem, resulting in the difficulty of how to efficiently combine the multiple cases into one solution for the new problem. This article investigates a new technique in complex adaptation. It presents the closest retrieved cases to a neural network, thus learning about the domain of the problem being solved. The new problem is then fed to the trained neural network and the output becomes the solution to that problem. The methodology is applied to a problem in steel construction, and the sought output is the cost estimation of pre-engineered steel buildings. Several experiments are conducted to prove that these steps are successful. System verification is done and shows that both the system and the methodology are successful in developing a complete adaptation mechanism in CBR.