Adaptation for Case-based Reasoning Classifier: An Extended Framework With Introspection Learning Approach
Hui Juan Zhao, Zejun Gong · 2023
The learning ability has a great influence on further application development and generalization ability of Case based reasoning (CBR), the traditional CBR framework could not satisfy the strict performance requirement of certain complex scenarios, While the improvement of CBR classifier framework is also a challenging task. Therefore, in order to ensure the classification performance, improve the leaning ability, expand the range of fields of application. An adapted 5R framework for CBR classifier is proposed: A new case reconstitute process is added between case retrieve and case reuse, where a reconstitution strategy with risk factor based on introspective learning is proposed to evaluate the Credibility of suggested solutions, then divide the target cases into different groups, and conduct the revise on the cases that are evaluated as risky to guarantee the reliability of the results. What's more a weight optimization based on evolutionary computation is also proposed to improve the retrieval accuracy. The comparison experiment with other common models, indicates that the proposed 5R framework could provide a more accurate and reliable result, verifies the feasibility of the framework, proves that the proposed model has better learning ability, which could use the potential information from the risk factor, improve the classification accuracy, and guarantee the reliability at the same time.