Study on Robust Granular Computing for Clustering and Regression and Implemented on Smart Phone Systems

• Main Authors: Nai-Cheng Shih, 施乃丞
• Other Authors: Jin-Tsong Jeng
• Format: Others
• Language: en_US
• Published: 2013
• Online Access: http://ndltd.ncl.edu.tw/handle/3mkyj5

碩士 === 國立虎尾科技大學 === 資訊工程研究所 === 101 === Clustering algorithms and regression analysis have been widely used in computer science, neural networks, machine learning and artificial intelligence etc for data analysis. At present, data format in the above fields has changed from dot data to symbolic data. Hence, how to deal with symbolic data become very important? In this thesis, we proposed interval rough robust granular computing for improving clustering and regression problem on symbolic data analysis. In general, Witold granular computing easily handled the symbolic data. However, if we consider the symbolic data that included noise and outliers, the Witold granular clustering or George Panoutsos’s granular clustering method will be effected via noise and outliers on the performance. Besides, noise and outliers are also effect the performance of granular regression box approach. Because granular regression box is easy effected via noise and outliers, it can’t search global optimization in the noisy environment. Hence, we proposed robust granular computing approach that used different distance measure approaches, rough sets theory and concept of determine annealing to improve granular clustering. At the same time, we applied the proposed approach to interval fuzzy clustering (IFC) and granular box regression (GBR). Because IFCM and GBR is difficult to decide initial structure for symbolic data, we applied the proposed rough robust granular computing to improve the drawback of IFCM and GBR for determining the initial structure under the symbolic data. Finally, we also implement the proposed algorithms to smart phone of Android systems.