An enhanced case-based reasoning model for supporting inference missing attribute and its feature weight

Case-based reasoning (CBR) methodology corresponds to problem solving by retrieving similar cases from a case base. In most research, the attribute values of each case are assumed to be complete. However, in reality, we encounter a situation that cases are collected from many different sources (e.g., books, papers) when a case base is built. Different sources with different attribute collections result in some null attribute values, which affect reasoning accuracy. The present study proposes an enhanced CBR model to improve accuracy when missing values occur in a case base. Association rules are applied to find the possible missing values, and rule's confidence is used to justify the feature weight. However, using the whole case base to mine the rules may involve the "noise" of different classifications, which affects the rules. To improve the quality of rules, classification is performed before rules are mined. The UCI Machine Learning Repository was used for evaluation to show the improved reasoning correctness with the proposed method. Our adapted CBR method has higher accuracy than unclassified and original case base reasoning.