臺灣博碩士論文加值系統

隨著資訊科技的成長，大部份的公司擁有大量的數位資料，例如：員工、客戶和交易資料…等等。因此，從這些資料探勘出有用的模型變成一個很重要的議題。學習向量量化是一個以代表點為基礎的分類技術並且可以在合理的時間內處理大量的資料。然而，由於傳統的學習向量量化是利用歐基理德距離所以僅能處理數值型資料而無法直接處理種類型資料，種類型資料必須進階使用1-of-k的方法做轉換才能處理。不過經由1-of-k轉換後，種類型資料會遺失原有的語義資訊並且導致分類效率下降。本篇研究，我們設計一個改良式學習向量量化藉由利用可以表達種類型資料之間關係的距離階層去處理混合型的資料。實驗的結果證明改良式學習向量量化在分類混合型資料可以優於傳統型的學習向量量化。

With rapid growth of information technology, most of corporations have collected a large amount of digital data, such as data regarding employees, customers and transactions, etc. Thus, mining useful patterns from the data becomes an important issue. Learning Vector Quantization (LVQ) is a prototype-based classification technique and can process a large volume of data within reasonable computation time. However, traditional LVQ process only numeric data due to the use of Euclidean distance but cannot directly handle categorical data which must be converted in advance, by typically using 1-of-k method. Nevertheless, after the conversion, categorical data lose their semantic information, leading to reduced classification performance. In this work, we propose an Improved LVQ (ILVQ) to deal with mixed-type data by using distance hierarchy for expressing relationship between categorical data. Experimental results prove ILVQ is better than traditional LVQ in classifying mixed-type data.

摘要 i
Abstract ii
1. Introduction 1
1.1 Motivation 1
1.2 Objective 1
1.3 Organization 2
2. Literature Review 3
2.1 Learning Vector Quantization 3
2.2 A generalized transform mechanism for categorical data 6
2.3 Distance Hierarchy 7
3. Improved Learning Vector Quantization 9
3.1 Improved Model with Distance Hierarchy 9
3.1.1 Training Mechanism on Improved LVQ 9
3.1.2 Adjustment on Distance hierarchy 10
3.1.3 Lowest Common Point Matrix for Adaptation 15
3.1.4 Consideration on Special Case for Adaptation 17
3.2 Construct Distance Hierarchy 18
3.3 The Overview of ILVQ 20
4. Experiments 21
4.1 Accuracy analysis on real mixed dataset 23
4.1.1 Adult 23
4.1.2 Contraceptive Method Choice 25
4.1.3 Australian 26
4.2 Accuracy analysis on real categorical dataset 28
4.2.1 Nursery 28
4.2.2 Car Evaluation 30
4.2.3 Hayes-Roth 32
4.3 Analysis of experimental results 33
4.4 Analysis on parameter sensitivity 35
4.4.1 Number of Codebook Vectors 35
4.4.2 Learning Rate 38
4.4.3 Training Time 40
4.5 Analysis on Reversed Distance Hierarchy 42
5. Conclusion 45
References 46

[1]C. C. Hsu, "Generalizing self-organizing map for categorical data," IEEE transactions on Neural Networks, vol. 17, pp. 294-304, 2006.
[2]C.-C. Hsu, S.-H. Lin, and W.-S. Tai, "Apply extended self-organizing map to cluster and classify mixed-type data," Neurocomputing, p. 11, 2011.
[3]C.-C. Hsu and S.-H. Lin, "Visualized Analysis of Mixed Numeric and Categorical Data Via Extended Self-Organizing Map," Neural Networks and Learning Systems, IEEE Transactions on, vol. 23, pp. 72-86, 2012.
[4]T. Kohonen, Self-organization and associative memory, 2nd ed. Berlin ; New York: Springer-Verlag, 1988.
[5]T. Kohonen, "Improved versions of learning vector quantization," in Neural Networks, 1990., 1990 IJCNN International Joint Conference on, 1990, pp. 545-550 vol.1.
[6]J. C. Bezdek and N. R. Pal, "Two soft relatives of learning vector quantization," Neural Netw., vol. 8, pp. 729-743, 1995.
[7]T. Kohonen, "Learning vector quantization," in The Handbook of Brain Theory and Neural Networks, M. A. Arbib, Ed., ed Cambridge, MA: MIT Press, 1995, pp. 537-540.
[8]T. Kohonen, J. Kangas, J. Laaksonen, and K. Torkkola, "LVQPAK: A software package for the correct application of Learning Vector Quantization algorithms," in Int’l. Joint Conf. on Neural Networks, 1995, pp. 725-730.
[9]P. Somervuo and T. Kohonen, "Self-Organizing Maps and Learning Vector Quantization for Feature Sequences," Neural Processing Letters, vol. 10, pp. 151-159, 1999.
[10]T. Kohonen, S. Kaski, K. Lagus, J. Salojarvi, J. Honkela, V. Paatero, and A. Saarela, "Self Organization of a Massive Document Collection," IEEE TRANSACTIONS ON NEURAL NETWORKS, vol. 11, 2000.
[11]T. Kohonen, Self-Organizing Maps vol. 30. New York: Springer, 2001.
[12]G. Das and H. M. a. P. Ronkainen, "Similarity of Attributes by External Probes," presented at the American Association for Artificial Intelligence, 1998.
[13]C. R. Palmer and C. Faloutsos, "Electricity based external similarity of categorical attributes," Lecture notes in computer science, pp. 486-500, 2003.
[14]C. J. Merz and P. Murphy. (1996). http://www.ics.uci.edu/~mlearn/MLRepository.html.
[15]J. Brownlee. (2007). http://wekaclassalgos.sourceforge.net/.