透過安裝於骨盆底肌肉訓練輔助器上的感測器，可以取得如施力、時間等資訊，分析這些資料並將其分群，提供給醫師判斷，賦予其專業認定的特性給各cluster。這些學習後的資料便成為患者的個人化參考依據，讓病人不在醫院時也能正確練習。
      本論文的目的在研究分析三類分群演算法，將其運用在骨盆底肌肉訓練的資料分群上。在centroid-based的部份引用了k-means，density-based的部份提出了源自DBSCAN的density-split，而connectivity-based則以SLINK的概念發展了chain。此外，針對noise造成的問題，本論文提出在分群前先排除noise的方法，欲藉此改善分群結果，讓醫師判讀與後續使用的參考資料更為精確。

      By attaching a pressure sensor on the device of pelvic floor muscle training (PFMT), we can collect data such as force and time. After data clustering, the proposed system can provide the PFMT data to the doctor. The doctor identifies the cluster by his/her professional knowledge. This identified data becomes the personal training data of the patient.
      The purpose of this thesis is studying three kinds of clustering algorithms, and implementing it for clustering the data of PFMT. In centroid-based, we reference "k-means". In density-based, we propose "density-split" which is inspired by DBSCAN. Finally, in connectivity- based, we propose "chain" which based on the concept of SLINK. Besides, in order to solve the problem caused by noise. We propose a method that excludes noise before clustering which can improve clustering result, and provide more accurate training data for clinical use.

目錄
第一章	緒論	1
1.1	研究背景	1
1.2	研究動機	1
1.3	研究目標	2
第二章	相關研究	5
2.1	Clustering	5
2.1.1	Centroid-Based	5
2.1.2	Density-Based	8
2.1.3	Connectivity-Based	10
2.2	Noise Detection	11
第三章	研究內容與方法	12
3.1	評估方法	12
3.2	測試資料	13
3.3	Centroid-Based : k-Means	14
3.4	Density-Based : Density-Split	17
3.5	Connectivity-Based : Chain	20
3.6	Noise處理	24
3.6.1	排除方法	24
3.6.2	配合Clustering的實作說明	26
第四章	實驗與結果討論	27
4.1	實驗說明	27
4.1.1	實驗設定	27
4.1.2	實驗環境	29
4.2	實驗結果	31
4.2.1	k-Means	31
4.2.2	Density-Split	34
4.2.3	Chain	37
4.2.4	實驗數據分析與討論	39
4.3	實作於PFMT的實際資料	42
4.3.1	實作說明	42
4.3.2	實作結果	44
4.3.3	討論與建議	46
第五章	結論與未來相關研究	47
5.1	結論	47
5.2	未來相關研究	48
參考文獻		49
附錄—英文論文	51

圖目錄
圖1.1  資料處理流程	4
圖2.1  k-Means步驟	6
圖2.2  Density-connected說明 (Source: [Ester 1996])	9
圖3.1  測試資料中的cluster基本類型	14
圖3.2  隨機初始centroid比較	14
圖3.3  Chain受noise影響圖	21
圖4.1  模擬資料類型	29
圖4.2  k-Means – 資料類型1	32
圖4.3  k-Means – 資料類型2	32
圖4.4  k-Means – 資料類型3	32
圖4.5  k-Means – 資料類型4	33
圖4.6  k-Means – 資料類型5	33
圖4.7  DS – 資料類型1	35
圖4.8  DS – 資料類型2	35
圖4.9  DS – 資料類型3	35
圖4.10  DS – 資料類型4	36
圖4.11  DS – 資料類型5	36
圖4.12  Chain – 資料類型1	37
圖4.13  Chain – 資料類型2	37
圖4.14  Chain – 資料類型3	38
圖4.15  Chain – 資料類型4	38
圖4.16  Chain – 資料類型5	38
圖4.17  三主要演算法比較執行結果	40
圖4.18  三主要演算法配合EN後比較執行結果	40
圖4.19  k-Means在原始實際資料的問題	42
圖4.20  DS在原始實際資料的問題	43
圖4.21  Chain在原始實際資料的問題	43
圖4.22  k-means – 實際資料	45
圖4.23  DS – 實際資料	45
圖4.24  Chain – 實際資料	45

表目錄
表4.1  PFMT裝置規格表	30
表4.2  實驗軟硬體表	30
表4.3  k-Means執行結果	33
表4.4  Density-Split執行結果	36
表4.5  Chain執行結果	39
表4.6  三主要演算法比較執行結果	39
表4.7  三主要演算法配合EN後比較執行結果	41

[Ester 1996] M. Ester, H.-P. Kriegel, J. Sander, X. Xu, “A density-based algorithm for discovering clusters in large spatial data sets with noise,” Proceedings of the Second International Conference on Knowledge Discovery and Data Mining, Portland, OR, pp. 226–231, 1996.

[Färber 2010] I. Färber, S. Günnemann, H.P. Kriegel, P. Kröger, E. Müller, E. Schubert, T. Seidl, and A. Zimek, “On Using Class-Label in Evaluation of Clusterings,”  1st International Workshop on Discovering,  Summarizing and Using Multiple Clusterings Held in Conjunction with KDD, 2010.

[Kashanian 2011] M. Kashanian, S. S. Ali, M. Nazemi, and S. Bahasadri, “Evaluation of the effect of pelvic floormuscle training (PFMT or Kegel exercise) and assisted pelvic floor muscle training (APFMT) by a resistance device (Kegelmaster device) on the urinary incontinence in women “comparison between them: a randomized trial”,” European Journal of Obstetrics Gynecology and Reproductive Biology, Vol. 159, No. 1, pp. 218–223, 2011.

[Lung 2010] Chung-Horng Lung, Chenjuan Zhou, “Using hierarchical agglomerative clustering in wireless sensor networks: An energy-efficient and flexible approach.” Ad Hoc Networks Vol. 8, Issue 3, pp, 328–344, 2010.

[MacQueen 1967] J. MacQueen, “Some methods for classification and analysis of multivariate observations,” Fifth Berkeley Symposium on Mathematics, Statistics and Probability, pp. 281–297, 1967.

[Maitra 2010] Ranjan Maitra, Anna D. Peterson, Arka P. Ghosh, “A systematic evaluation of different methods for initializing the K-means clustering algorithm,”  IEEE Transactions on Knowledge and Data Engineering, 2010.

[Onoda 2010] Takashi Onoda, Miho Sakai, Seiji Yamada, “Careful Seeding Based on Independent Component Analysis for k-Means Clustering,” 2010 IEEE/WIC/ACM International Conference on Web Intelligence and Intelligent Agent Technology, Vol.3, pp.112-115, 2010.

[Rand 1971] William M. Rand, “Objective criteria for the evaluation of clustering methods,”  Journal of the American Statistical Association, Vol. 66, No. 336, pp. 846-850, 1971.

[Romero 2010] Jose Diaz-Romero, Salvatore Romeo, Judith V.M.G. Bovée, Pancras C.W. Hogendoorn, Paul F. Heini, Pierre Mainil-Varlet, “Hierarchical clustering of flow cytometry data for the study of conventional central chondrosarcoma,” Journal of Cellular Physiology Vol. 225, Issue 2, pp. 601–611, 2010.

[Williams 2002] G. Williams, R. Baxter, H. He, S. Hawkins and L. Gu, “A Comparative Study for RNN for Outlier Detection in Data Mining,” Proceedings of the 2nd IEEE International Conference on Data Mining, pp. 709, 2002.

[Xiao 2012] X. Xiao, S. Cao, L. Huang and Y. Tang, “Scientific Communities Found Based on the Path Structure of Citation Network,” International Journal of Intelligence Science, Vol. 2 No. 1, pp. 16-21, 2012.

[Zhong 2011] Caiming Zhong, Duoqian Miao, Pasi Franti, “Minimum spanning tree based split-and-merge: A hierarchical clustering method,” Information Sciences, Vol. 181, Issue 16, pp.3397–3410, 2011.