클러스터링 성능 향상을 위한 실루엣 기반 시뮬레이티드 어닐링

A Silhouette-Based Simulated Annealing for Improving Clustering Performance

Data clustering is one of the most important and difficult technique in data mining. K-means is popular and efficient data clustering method based on intra-cluster distance valid index. This is sensitive for initialization and has the possibility to be stuck in local optimum because of hill climbing clustering method. We need a heuristic data clustering method because clustering can be formally considered as a particular kind of NP-hard grouping problem. The heuristic algorithm is also not computationally feasible in practice, especially when using Silhouette valid index for data clustering large datasets with large number of clusters. Therefore, we need a robust and efficient heuristic clustering method to find the global optimum (not local optimum) within limited computation time. The objective of this paper is to propose machine learning novel simulated annealing (NSA) based on Silhouette to find the global optimal solution starting good initial solution using probabilistic relative distance rate. NSA is validated using UCI machine learning repository datasets comparing to swarm intelligence methods which are Group Search Optimization (GSO), Artificial Bee Colony (ABC), and Biogeography Based Optimization (BBO) by simulation experiments and analysis. Our proposed NSA has significantly improved accuracy and efficiency to find the global optimal data clustering solution for complex and costly data mining process.