A Literature Review of Expert Problem Solving using Analogy

We consider software project cost estimation from a problem solving perspective. Taking a cognitive psychological approach, we argue that the algorithmic basis for CBR tools is not representative of human problem solving and this mismatch could account for inconsistent results. We describe the fundamentals of problem solving, focusing on experts solving ill-defined problems. This is supplemented by a systematic literature review of empirical studies of expert problem solving of nontrivial problems. We identified twelve studies. These studies suggest that analogical reasoning plays an important role in problem solving, but that CBR tools do not model this in a biologically plausible way. For example, the ability to induce structure and therefore find deeper analogies is widely seen as the hallmark of an expert. However, CBR tools fail to provide support for this type of reasoning for prediction. We conclude this mismatch between experts’ cognitive processes and software tools contributes to the erratic performance of analogy-based prediction