Abstract
Case-Based Reasoning (CBR) system relies on reuse for solving new problems. The system uses the experiences it previously acquired and stored into its case base to address the newly faced problems. A static and non-evolutive case base hinders the system and limits the accuracy of the CBR in problem-solving. While a massive case base can affect the resolution time. Randomization represents a way to generate data without deteriorating the spatial image of the case base and by extension the search time as well. However, the cases generated by randomization are not necessarily valid and require a thorough validation process to access their validity. This paper presents a new amplification technique based on randomization for a CBR system incorporating a structured case-base that speeds up case retrieval while supporting case retention. The generated data by randomization is validated through a three-layer validation process: coherence verification, stochastic validation, and absolute validation. Furthermore, we propose a new way to segment the case base along with new similarity functions based on features’ weights to speed CBR retrieval. We carried out experiments on mammography mass and thyroid disease datasets to validate our approach, where the proposed approach is compared to several popular supervised machine-learning methods and other related works that utilize the same datasets. Experiments have shown that our approach can generate relevant data, which significantly improves the resolution accuracy and makes CBR a good competitor to classification methods.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call