目前企業各方面決策幾乎是以歷史資料探勘(Data mining)結果分析為基礎，故資料庫的完整性則十分重要，若是資料庫中出現過多的遺漏值(Missing value)，則容易使探勘結果可靠性降低，因此遺漏值推估問題成為許多研究者努力的目標。在早期針對遺漏值推估問題已有許多方法如直接刪除、以平均或眾數回填等，但這樣的方法若遇到含有大量的遺漏值資料時，通常無法提供最後決策有利參考資訊。 本研究在搜尋大量文獻及經過多方討論後，決定改良粒子群最佳化演算法( Particle swarm optimization,PSO)提出整合反彈機制、K-means及PSO之反彈機制KPSO( hybrid k-means and efficient Particle swarm optimization in clustering ,RKPSO)，對資料作分群後以群內平均回填該群遺漏值欄位，並與文獻結果比較及多種資料庫作實驗證實本研究方法之可行性。 Nowadays, enterprises’ decisions almost are according to the analytic outcomes of past data mining. Hence, it’s quite important to keep the solidity of database. In early days , people always used the average、mode to backfill missing value or directly delete the data that includes missing values. However, it’s not a very good way to solve this problem. If there are too many missing values within the database it can not offer reliable information. After searching lots of conferences and discussing with my professor, we try to integrate K-means、efficient in PSO to cluster the data and estimate missing values. At last, we found RKPSO (hybrid k-means and efficient Particle swarm optimization in clustering) better than other ways on through actual database experiment and conferences comparison.
