Case-based reasoning (CBR) is a problem-solving and reasoning paradigm that can overcome limitations of the rule-based expert systems. Instead of rules, a CBR system stores and maintains past cases as a case base. When a new problem arises, the system searches through the case base for similar cases to construct a solution for addressing the new problem. Nearest neighbor is a common CBR algorithm for retrieving similar cases, whose similarity function is sensitive to irrelevant attributes. To ensure the effective retrieval of similar cases, a hybrid case-based reasoning approach which employs statistical evaluation for automatically assigning attribute weights and nearest-neighbor algorithm for case retrieval is proposed. This approach is applied to business failure prediction in Australia. The results indicate that in this case it outperforms discriminant analysis in terms of classification accuracy and is an effective and competitive alternative in providing early warnings of those companies at risk of failing in a comprehensible manner.
展开▼
