在研究所的生活中，首先要感謝的是我的指導教授李育杰老師在這段期間的指教。不
論是做學問時應具備的態度極思考方式，以及實際生活面待人處事之道理。在這段期
間所學到的知識以及研究態度，對我往後的人生必定受用無窮。此外，在此要特別感
謝百忙之中抽空前來參與口試的各位委員們: 葉苡任教授、鮑興國教授以及黃文瀚教
授，諸位都提供了我寶貴的意見，從不同的角度切入並點出我研究中的盲點，協助我
將論文內容變得更加完善。
同時，也要特別感謝實驗室的研究夥伴們，感謝勤彥學長協助論文撰寫；茲文學
長、信融和宜恩一起討論論文內容；學弟妹宥樺、采瀅、哲豪及威志讓實驗室氣氛變
得更加活絡。在實驗室的各位之照顧下，這兩年的生活過得十分充實。當然，隔壁實
驗室的各位同胞們，也感謝你們改善我們實驗室的氣氛。無論是在學業或是生活上，
大家能夠在一起互相支持、鼓勵，共同學習、成長，令我十分感激。
最後，僅以此論文獻給我的家人們，感謝你們對我的包容，即使很少有時間待在
家中，各位也盡可能的配合我的時間去安排家庭活動。若非你們的支持，此篇論文將
難以付梓。本人在此致上最高的敬意，感謝各位!

The time series classi cation problem has been studied over a decade. Since time series
is a ubiquitous type of data, there has been much e ort devoted to this issue. Basically,
time series classi cation approaches can be categorized into three types, including
distance-based, model-based, and feature-based. In this paper, we focus on feature-based
methods, which represent time series into a set of characterized values. However, features
generated by most of existing representation techniques are not completely interpretable.
Due to this fact, a novel time series representation, envelope, is proposed. This supervised
feature extraction method transforms time series into simple 1=0= ? 1 values. It is
always important to nd the most discriminating features for data mining tasks. Hence,
a heuristic is introduced for determining the best representation. Moreover, this new
representation enjoys the characteristic of sparsity which is essential property for applying
compressed sensing. With this advantage, we can compress data sets and reduce
model complexity. Furthermore, the transformed features are interpretable via visualization.
Envelope shows the shape of time series and de nes the similarity between time
series. We demonstrate that this representation provides comparable result on numerous
benchmark datasets.

[1] L. Atzori, A. Iera, and G. Morabito. The internet of things: A survey. Comput.
Netw., 54(15):2787{2805, Oct. 2010.
[2] R. Calderbank, S. Jafarpour, and R. Schapire. Compressed learning: Universal sparse
dimensionality reduction and learning in the measurement domain.
[3] E. Cand es and M. Wakin. An introduction to compressive sampling. IEEE Signal
Processing Magazine, 25(2):21{30, 2008.
[4] C.-C. Chang and C.-J. Lin. Libsvm: A library for support vector machines. ACM
Trans. Intell. Syst. Technol., 2(3):27:1{27:27, May 2011.
[5] Y. Chen, E. Keogh, B. Hu, N. Begum, A. Bagnall, A. Mueen, and G. Batista. The
ucr time series classi cation archive, July 2015. www.cs.ucr.edu/~eamonn/time_
series_data/.
[6] H. Ding, G. Trajcevski, P. Scheuermann, X. Wang, and E. Keogh. Querying and
mining of time series data: Experimental comparison of representations and distance
measures. Proc. VLDB Endow., 1(2):1542{1552, Aug. 2008.
[7] T.-c. Fu. A review on time series data mining. Eng. Appl. Artif. Intell., 24(1):164{
181, Feb. 2011.
[8] P. Geurts. Pattern extraction for time series classi cation. In L. De Raedt and
A. Siebes, editors, Principles of Data Mining and Knowledge Discovery, volume 2168
of Lecture Notes in Computer Science, pages 115{127. Springer Berlin Heidelberg,
2001.
[9] J. Gubbi, R. Buyya, S. Marusic, and M. Palaniswami. Internet of things (iot): A
vision, architectural elements, and future directions. Future Gener. Comput. Syst.,
29(7):1645{1660, Sept. 2013.
[10] E. Keogh and J. Lin. Hot sax: E ciently nding the most unusual time series
subsequence. pages 226{233, 2005.
[11] J. R. Kwapisz, G. M. Weiss, and S. A. Moore. Activity recognition using cell phone
accelerometers. SIGKDD Explor. Newsl., 12(2):74{82, Mar. 2011.
[12] J. Lin, E. Keogh, S. Lonardi, and B. Chiu. A symbolic representation of time series,
with implications for streaming algorithms. In Proceedings of the 8th ACM SIGMOD
Workshop on Research Issues in Data Mining and Knowledge Discovery, DMKD '03,
pages 2{11, New York, NY, USA, 2003. ACM.
[13] J. Lines, L. M. Davis, J. Hills, and A. Bagnall. A shapelet transform for time series
classi cation. In Proceedings of the 18th ACM SIGKDD International Conference on
Knowledge Discovery and Data Mining, KDD '12, pages 289{297, New York, NY,
USA, 2012. ACM.
[14] H. Ocak. Automatic detection of epileptic seizures in eeg using discrete wavelet
transform and approximate entropy. Expert Syst. Appl., 36(2):2027{2036, Mar. 2009.
[15] K. Polat and S. Gune s. Classi cation of epileptiform eeg using a hybrid system
based on decision tree classi er and fast fourier transform. Applied Mathematics and
Computation, 187(2):1017{1026, 2007.
[16] C. A. Ratanamahatana, E. J. Keogh, A. J. Bagnall, and S. Lonardi. A novel bit
level time series representation with implication of similarity search and clustering.
In Advances in Knowledge Discovery and Data Mining, 9th Paci c-Asia Conference,
PAKDD 2005, Hanoi, Vietnam, May 18-20, 2005, Proceedings, pages 771{777, 2005.
[17] N. Ravi, N. D, P. Mysore, and M. L. Littman. Activity recognition from accelerometer
data. In In Proceedings of the Seventeenth Conference on Innovative Applications of
Arti cial Intelligence(IAAI, pages 1541{1546. AAAI Press, 2005.
[18] Z. Xing, J. Pei, and E. Keogh. A brief survey on sequence classi cation. SIGKDD
Explor. Newsl., 12(1):40{48, Nov. 2010.
[19] L. Ye and E. Keogh. Time series shapelets: A new primitive for data mining. In
Proceedings of the 15th ACM SIGKDD International Conference on Knowledge Dis-
covery and Data Mining, KDD '09, pages 947{956, New York, NY, USA, 2009. ACM.