一個針對二維資料的機率式重新分配模擬退火分群法的介面

Abstract

分群法是一個被廣泛使用的自動化資料分類技術，常用於機器學習或資料探勘的先置處理步驟前置。其中模擬退火法是一種模擬多粒子物理系統由較高能量熱平衡狀態降溫到較低能量熱平衡狀態過程以取得最少TSSE的分群法。SAKM-clustering整合了模擬退火法取得系統最小能量及 K-means 快速搜尋的能力。它適用於搜尋多維特徵空間中適當分群並使得分群結果在相似度測度上最佳化。在本論文中我們針對二維資料實作了一個界面以觀察SAKM-clustering及K-means的分群過程效能比較。

Clustering is an extensively used technique for automatic data classification, such as in the preprocessing of machine learning and data mining. Simulated-annealing is a clustering technique, which obtains the minimum TSSE of a group of data by simulating the cooling down process of a many-particle physical system from a state in thermal equilibrium with higher energy to another state in thermal equilibrium with lower energy. The SAKM-Clustering integrates the power of simulated-annealing for obtaining minimum energy configuration and the searching capability of K-means algorithm. It is used to search proper clusters in multidimensional feature space such that a similarity metric of the resulting clusters is optimized. In this thesis we implement a GUI to observe the clustering processes and to compare the performances of the SAKM-clustering as will as the K-means clustering.