本論文提出一個限制推理型之粒子群與基因演算法用以解決資料探勘之分類問題。由於粒子群與基因演算法主要是透過適應值函數來評估分類規則,但是當問題的複雜度提高時,除了搜尋時間會相對提高外,且有可能會陷入區域最佳解之困境。基於上述之缺點,本研究整合限制推理及粒子更新機制於粒子群演算法架構中,用來減少粒子產生的搜尋空間,使得粒子群演算法能夠更快速的找出符合限制之最佳解。此外本研究提出混合式粒子更新機制中,結合粒子群與基因演算法更新機制以補足粒子群演算法於區域搜尋能力不足的缺點,使粒子之搜尋範圍及效能得到良好的提升。根據實驗結果顯示,限制式粒子群及基因演算法較一般傳統的粒子群演算法有更快速且較符合目標之結果。
This paper aims to develop a constraint-based particle swarm optimization and genetic algorithm approach for mining classification rules. Existing particle swarm optimization (PSO) designed for rule induction evaluates the rules as whole via a fitness function. Major drawbacks of PSO for rule induction include computation inefficiency, accuracy and rule expressiveness. In this paper, we propose a constraint-based particle swarm optimization and genetic algorithm(CBPSOGA) approach for mining classification rules.This approach allows constraints to be specified as relationship among attributes according to predefined requirements of user’s preference in the form of a constraint network. Additionally, the new update approach based on genetic algorithm (GA) is incorporated to produce better classification rules. The proposed approach is compared with a regular PSO,constraint-based GA and constraint-based PSO algorithms using UCI repository data sets. Better classification accurate rates form CBPSOGA are demonstrated.