在案例推理(CBR)研究中,主要應透過案例推理機制找出最符合現行問題需求之結果。在本研究使用機器學習的技術包括基因規劃法(GP)及基因演算法(GA)來產生CBR tree以提升單一CBR案例推理系統之學習能力。基因規劃法用於建立案例切割規則來區隔各子案例庫;基因演算法用以求出各子案例庫下特徵屬性比對相似值的對應權重。本研究乃利用UCI及DELVE資料庫中,包含分類及推估問題共六組實驗資料進行模式建構,同時也應用GA-CBR 、C5.0及CART三種技術進行實驗並比較其學習成效。實驗結果顯示CBR tree效果明顯比其它三種技術優越。
A critical issue must be correctly recognized in case-based reasoning (CBR) that is to retrieve not just a similar historical case but a usefully similar case from case base to the target problem. For this reason, this paper examines application of machine learning techniques, genetic programming (GP) and genetic algorithm (GA) to the integration of domain knowledge into the CBR. We call the integrated system GP-Based CBR tree system in this paper. First, we apply classification problems based on GP for case base and split up into several subcase bases in accordance with some important features from case base. Second, GA would be used to determine weight sets for features similarity degree of several subcase bases, and we call this part GA-CBR. The CBR tree reaps the benefits of three systems. The hybrid approach combined GP and GA both techniques with CBR systems for increasing the overall accuracy. We experimentally assess six datasets and accuracy of generated was comparable to four approaches that were CBR tree, GA-CBR, C5.0 and CART over six datasets.