本研究改良傳統之類免疫系統,並在演算學習過程中導入機率性之概念;將分類問題分兩部份執行,第一部份採用類免疫系統之學習架構,找出資料或數據之最佳分群中心,將資料做分群,此部分為非監督式學習;第二部份加入機率之概念,即利用多變量高斯混合模型配適出每一類別之分類模型,並決定每一類別之多變量高斯混合模型中所含高斯分配之權重及個數;研究結果顯示,改良之機率性類免疫分類演算法(Probabilistic Artificial Immune Algorithm of Classification;簡稱PAIAC)在Iris及肝功能失調這兩個分類問題皆有不錯的表現;最後並以K-NN、BPN及SVM進行比較,在整體分類精確度的表現上PAIAC與BPN相近,且較SVM及K-NN為佳;PAIAC在參數設定、分群及分類問題方面相較於BPN及SVM之下皆較簡易、有彈性,因此在未來面臨有關資料之分群或分類問題時,PAIAC亦是不錯之選擇。
In this research, a probabilistic model is incorporated in the artificial immune of classification algorithm named as probabilistic Artificial Immune Algorithm of classification. The process of classification is in a two-stage execution. The first stage is determined the best cluster center in data and the number of clusters. This part is categorized as non-supervising learning. The second stage takes multivariate gauss mixture model to generate classification models precisely for each category, meanwhile weight of each gauss of model and number of assigning are determined for each category. Numerical study shows that PAIAC (Probabilistic Artificial Immune Algorithm of Classification, PAIAC) performs better than traditional artificial intelligence clustering method in the real cases of Iris and liver function study. A comparison among K-NN, BPN, SVM and PAIAC is conducted. It shows that the PAIAC makes no difference with BPN in accuracy but even batter than SVM and K-NN. In addition, the advantages of PAIAC include the easy configuration of parameters and higher flexibility. We recommend that the PAIAC is another choice when study the clustering or classification problems.