An Information-Based Approach to Compute Similarity Between Engineering Changes

Abstract

Utilizing past Engineering Change (EC) knowledge to predict the impact of a proposed EC effect requires an approach for computing similarity between ECs. This paper presents an approach for computing the similarity between ECs each defined by a set of disparate attributes. Since the available information is probabilistic, measures of information are used for defining measures to compute similarity between two attribute values or ECs. The semantics associated with attribute values are used to compute similarity between them. The similarities between attribute values are aggregated to compute the similarity between ECs in the context of overall goal. An example EC knowledge-base is used for evaluating our approach against a statistical approach and two state-of-the-art approaches, namely, metric space and probability-based. The evaluation is done from two perspectives: precision in retrieving similar ECs and success in predicting the impact. The results show that there is a statistically significant improvement in precision and success rate using our approach as compared to those using other approaches. In addition, based on the results, it can be inferred with 90% confidence that for a large number of changes (N >; 100) the success in predicting impact using our approach shall be greater than that obtained using the two state-of-the-art approaches.