隨著資訊科技的進步與電腦運算能力的提升,現今已有許多問題仰賴電腦輔助解決問題。然而分類問題在現實社會中佔有相當大的比例,因此分類問題確實值得深度探討。過去解決分類問題中,大多數演算法或模型可帶來不錯的效果,但對於眾多分類問題的求解穩定性卻不進理想。近年來啟發式演算法設計在許多問題上皆可得到相當不錯的效果,其主要是以演化的方式來學習問題本身的特性,以迭代的搜尋期望找尋可行的解決方案。本研究希望透過演化式演算法中的類免疫演算法為基礎,發展出一個類免疫分類演算法來解決分類問題,演算法包含克隆選擇、免疫抗體分級制度及新抗體繁殖,其中產生新抗體的方法將利用漸進式學習法達成,透過這些設計期望提高整體演算法的效率及收斂性。本研究實驗設計以網路異常入侵及信用評核為主要解決之分類問題,而實驗結果顯示本研究所提之類免疫演算法結合漸進式學習可在不同分類問題上得到相當好的分類效果。相較於其他分類演算法,本研究所建立之類免疫分類演算法具有相當程度的正確性及穩定性。
With the advance of information technology and the improvement of computing power, more and more problems has been solved and relied by the computer nowadays. Nevertheless the classification problems have been accounted for a sizeable proportion in the reality, therefore, it is worth a further immersing to this topic. In the past, classification problems that solved by the algorithm or model end up with a good result are often seen. However, the stability of the solution for many classification problems is not ideal enough. In recent years the design of the heuristic algorithm on many issues can be quite good results. It's mainly based on the way of evolution that learns the characteristics of the problem itself and expects to use iterative search to find a viable solution. This study hopes to construct a method based on a kind of evolutionary algorithms, artificial immune system, to develop an artificial immune classification algorithm and solve classification problems. The algorithm consists of clonal selection, antibody grading mechanism and new antibody expansion mechanism, among the use of incremental learning method to evolve new antibody and through these designs, expecting the improvement of overall efficiency and convergence of the algorithm. In this study, experimental design is mainly to solve intrusion detection and credit approval problems. The experimental results show that the proposed artificial immune system combined with incremental learning brings out positive results in different classification problems. The artificial immune classification algorithm in this study compared with other classification algorithms, the proposed method with a considerable degree of accuracy and stability.