臺灣博碩士論文加值系統

資料採掘(data mining)乃指以統計與數學等技術試圖從大量的資料堆中掘取有意義的模式(pattern)，並分析其趨勢(trend)以及相關性(correlation)的程序。有別於其它需要使用者事先假設資料間的關連性再用分析工具驗證其假設之法，資料採掘利用相似測度法及模式配對(pattern-matching)的技術;如規則基底分析(rule-based analysis)、類神經網路(artificial neural networks)、模糊邏輯(fuzzy logic)、K-最近類近法(K-nearest neighbor)、基因演算法(genetic algorithm)、高等視覺工具以及主成份分析(principal component analysis, PCA)等，以決定資料集內元素之重要關連性。群聚分析(cluster analysis)在資料採掘問題上扮演經常性的角色，它們往往是資料採掘分析上的首要工作。其主要目的在於確立資料之特徵值，依據其相似性而加以細分成若干群聚。
本論文主要針對時空性資料的群聚分析問題來發展適當的資料採掘技術，並將以空氣懸浮粒指數(PM10)分析為個案研究，因其能粹取動態之特徵與模式以提供研究季節性(時間)與地理性(空間)之變異性。然而，群聚分析之結果對輸入資料之尺度相當敏感。這意味著分析的結果可能隨時、日、週、月、季或年的資料而不同。如何選擇適當的時間尺度，實際上視各種應用目的而定。本論文亦以連結主義者(connectionist)方法(類神經網路為其主要代表)配合多重尺度分析(Multiresolution Analysis)，以探討時空性資料之相關性。於此，吾人將著重於自我映射組織圖(self-organizing map, SOM)模式與多重尺度小波轉換之研究，因其學理與應用正被廣泛地探討，更重要的是其能反映欲分析之資料的地理性(geographical)組織關係。基於此，吾人預期能夠去除以單一較小輸入尺度下之小區域特徵值或改善以單一較大尺度下之過度平滑之區域。關於實驗，我們將應用多重尺度連續性波紋轉換法，做為空氣懸浮粒之群聚分析的前端處理程序，再配合自我映射組織圖模式探討在不同長短時間尺度上對空間變化性之影響。本研究成果之應用相當廣泛，主要在於時空性資料中重要特性之萃取，例如，從遙測影像監控環境之改變與海面溫度之趨勢分析以進行地球環境改變之研究。

Abstract.....................................................I
Contents....................................................II
List of Figures and Tables..................................IV
Chapter 1 Introduction......................................1
Chapter 2 Multiscale Analysis Using Wavelet Transform.......5
2.1 Continuous wavelet transform........................5
2.2 Location and scale families.........................6
2.3 Discrete wavelet transform..........................7
2.4 The Pyramid Algorithm..............................11
2.5 Multiresolution analysis...........................15
2.6 Time-scale analysis................................16
Chapter 3 Self-Organizing Map Neural Networks..............20
3.1 Network structure..................................20
3.2 Learning algorithm.................................23
3.3 Visualization......................................27
3.4 Cluster analysis by modified self-organizing maps..31
Chapter 4 Computer Simulations On EPA Data.................34
4.1 Data description...................................34
4.2 Data preparation...................................34
4.3 Missing value removal..............................40
4.4 The scale issue in mining EPA data.................41
4.5 Performance analysis of clustering results.........66
Chapter 5 Conclusions.......................................81
Reference...................................................82

[1] David Donoho, and Hong-Ye Gao, “Wavelet analysis,” IEEE spectrum October 1996, pp26-35.
[2] Los Alamitors, and Los Vaqueros, “An introduction to wavelets”, IEEE Computational Science and Engineering, Summer 1995, vol. 2, num. 2.
[3] Esa Alhoniemi, Johan Himberg, Kimmo Kiviluoto, Jukka Parviainen, Juha Vesanto. (1997). Wavelet Toolbox of Matlab User’s Manual, Version 1.0beta, The MathWorks, Inc.
[4] Bruce, A. and Gao, H.-Y. (1994). S+Wavelets User’s Manual, Version 1.0, StatSci Division, MathSoft Inc., Seattle, WA.
[5] Shubha Kadambe, and G.F. Boudreaux-Bartels, “A comparison of a wavelet functions for pitch detection of speech signals,” 1991 IEEE.
[6] Fionn Murtagh of University of Ulster, “Wedding the wavelet transform and multivariate data analysis.”
[7] Olivier Rioul ,and martin Vetterli, “Wavelets and signal processing,” IEEE spectrum, October 1991.
[8] Robi Polikar, “The wavelet tutorial part I~IV.”
[9] Samuel Kaski of Helsinki University of Technology Neural Networks Research Centre, “Data Exploration using Self-Organizing Maps”, FIN-02150 Espoo 1997.
[10] A prochazka, M Storek, “Wavelet transform use for signal classification by self-organizing neural networks”, ‘Artificial Neural Networks’, 26-28 June 1995, Conference Publication No.409, cIEE, 1995.
[11] J. Noonan, “Fuzzy Kohonen Network for the classification of transients using the wavelet transform for feature extraction,” Elsevier Science Inc. 1995.
[12] Esa Alhoniemi, Johan Himberg, Kimmo Kiviluoto, Jukka Parviainen, Juha Vesanto. (1997). SOM Toolbox of Matlab User’s Manual, Version 1.0beta, The MathWorks, Inc.
[13] Sheng-Tun Li, Jeng-Jong Pan., and Shuzheng Cong, “Multiscale Study of the spatial variability in the cluster analysis of rainfall,” International Journal of Climatology, in July 9, 1997.
[14] Melody Y. Kiang, Uday R. Kulkarni, and Kar Yan Tam, “Self-organizing map network as an interactive clustering tool- an application to group technology,” Decision Support Systems 15(1995), pp. 351-374.
[15] Tatsuya Nomura, and Tsutomu Miyoshi, “An adaptive rule extraction with the fuzzy self-organizing map and a comparison with other methods,” 1995 IEEE proceedings of ISUMA-NAFIPS’95.
[16] Juha Vesanto, “ Data mining techniques based on the self-organizing map,” Espoo, Finland 26th MAY 1997.
[17] Sk Chen, P mangiameli, and D west, “ The comparative ability of self-organizing neural networks to define cluster structure,” Omega,Int. Mgmt Sci. Vol. 23, pp271-279,1995.
[18] Paul Mangiameli, Shaw K.Chen, and David West, “ A comparison of SOM neural network and hierarchical clustering methods,” European Journal of Operational Research 93 (1996) pp. 402-417.
[19] Teuvo Kohonen, “Self-Organizing Maps,” Second Edition of text book Published by Springer Inc, 1997.
[20] Carlos Serrano-Cinca, “ Self-organizing neural networks for financial diagnosis,” Decision Support Systems 17 (1996), pp. 227-238.
[21] B. Martin-del-Brio and N. Medrano, Feature Map Architectures for Pattern Recognition: Techniques for Automatic Region Selection, in: D. W. Pearson, N.C. Steele and R.F. Albrecht, Eds., Artificial Neural Nets and Genetic Algorithms (Springer-Verlag, Berlin, 1995) 124-127