城市交通日漸繁忙，事故發生機率亦隨之上升，過去有許多研究應用機器學習演算法，預測未來交通事故熱區、車流量或平均車速，然而多數的研究著重於如何提出新穎的機器學習架構。本論文的研究是以桃園市部分行政區為實驗場域，於各行政區取道路較密集的區域，每區域約16 平方公里，利用經典機器學習模型：CNN 與LSTM 組合訓練模型，分析在何種CNN-LSTM 層數組合下，能夠以較低的訓練成本，得到準確率較高的模型組合。

Urban traffic is getting busy, and the probability of accidents is also rising. In the past, many studies applied machine learning algorithms to predict hot spots of traffic accidents, traffic flow, or average speed. However, most of the research focused on how to propose novel machine learning architectures. The study in this thesis uses some regions of Taoyuan City as experimental fields. By taking dense road areas in these regions, each
area is about 16 square kilometers. This study uses the classic machine learning model: Convolution Neural Network(CNN) and Long Short-Term Memory(LSTM). In sensitivity to varied number of CNN-LSTM layers, this study examines the performance of higher accuracy and lower training cost.

[1] “中華民國 107 年桃園市警政統計年報,” accessed Jun 22, 2020. [Online].
Available: https://www.typd.gov.tw/uploads/files/tyhp100/107%E5%B9%
B4%E5%A0%B1.pdf
[2] Q. Zhu, J. Chen, D. Shi, L. Zhu, X. Bai, X. Duan, and Y. Liu, “Learning temporal
and spatial correlations jointly: A unified framework for wind speed prediction,”
IEEE Transactions on Sustainable Energy, vol. 11, no. 1, JAN 2020, pp. 509–523.
[3] C. Olah, “Understanding lstm networks,” accessed Jun 17, 2020. [Online]. Available:
colah.github.io/posts/2015-08-Understanding-LSTMs/
[4] A. Karpathy, “The unreasonable effectiveness of recurrent neural networks,”
accessed Jun 18, 2020. [Online]. Available: http://karpathy.github.io/2015/05/21/
rnn-effectiveness/
[5] D. Zang, J. Ling, K. Tang, and J. Cheng, “Long-term traffic speed prediction based
on multiscale spatio-temporal feature learning network,” IEEE Transactions on
Intelligent Transportation Systems, vol. 20, no. 10, Oct 2019, pp. 3700–3709.
[6] C. Chen, X. Fang, C. Zheng, L. Xiao, M.Cheng, and C.Wang, “Sdcae: Stack denoising
convolutional autoencoder model for accident risk prediction via traffic big data,” in
Proceedings of Sixth International Conference on Advanced Cloud and Big Data,
2018.
[7] Y. He, Z. Liu, X. Zhou, and B. Zhong, “Analysis of urban traffic accidents features
and correlation with traffic congestion in large-scale construction district,” in
Proceedings of 2017 International Conference on Smart Grid and Electrical
Automation (ICSGEA), 2017, pp. 641–644.
[8] V. Milanés, S. E. Shladover, J. Spring, C. Nowakowski, H. Kawazoe, and M. Nakamura,
“Cooperative adaptive cruise control in real traffic situations,” IEEE Transactions
on Intelligent Transportation Systems, vol. 15, no. 1, pp. 296–305, 2014.
[9] Y. Murat, Fuzzy Clustering Approach for Accident Black Spot Centers
Determination, Fuzzy Logic-Emerging Technologies and Applications, 03 2012.
[10] A. Koesdwiady, R. Soua, and F. Karray, “Improving traffic flow prediction with
weather information in connected cars: A deep learning approach,” IEEE Transactions
on Vehicular Technology, vol. 65, no. 12, pp. 9508–9517, 2016.
[11] “五分鐘學 gis，時空資料視覺化表達,” accessed Jun 23, 2020. [Online].
Available: https://codertw.com/%E7%A8%8B%E5%BC%8F%E8%AA%9E%E8%
A8%80/511780/
[12] W. Wei, H. Wu, and H. Ma, “An autoencoder and lstm-based traffic flow prediction
method,” Sensors, vol. 19, p. 2946, 07 2019.
[13] Q. Chen, X. Song, H. Yamada, and R. Shibasaki, “Learning deep representation
from big and heterogeneous data for traffic accident inference,” in Proceedings of
the Thirtieth AAAI Conference on Artificial Intelligence, ser. AAAI’16. AAAI
Press, 2016, p. 338–344.
[14] A. Krizhevsky, I. Sutskever, and G. E. Hinton, “Imagenet classification with deep
convolutional neural networks,” in Proceedings of the 25th International
Conference on Neural Information Processing Systems - Volume 1, ser. NIPS’12.
Red Hook, NY, USA: Curran Associates Inc., 2012, p. 1097–1105.
[15] M. Đalto, “Deep neural networks for time series prediction with applications in ultrashort-
term wind forecasting,” in Proceedings of Rn(01), vol.1, 02 2014.
[16] “Research on text classification based on cnn and lstm,” in Proceedings of 2019
IEEE International Conference on Artificial Intelligence and Computer Applications
(ICAICA), Y. Luan and S. Lin, Eds., 2019, pp. 352–355.
[17] S. Hochreiter and J. Schmidhuber, “Long short-term memory,” Neural
Computation, vol. 9, 12 1997, pp. 1735–80.
[18] Y. Heryadi and H. L. H. S. Warnars, “Learning temporal representation of transaction
amount for fraudulent transaction recognition using cnn, stacked lstm, and cnnlstm,”
in Proceedings of 2017 IEEE International Conference on Cybernetics and
Computational Intelligence (CyberneticsCom), 2017, pp. 84–89.
[19] “深度學習：使用激勵函數的目的、如何選擇激勵函數 deep learning : the
role of the activation function,” accessed Jun 24, 2020. [Online]. Available:
https://mropengate.blogspot.com/2017/02/deep-learning-role-of-activation.html
[20] M. Nielsen, “Why are deep learning networks hard to train?” accessed Jun 18,
2020. [Online]. Available: http://neuralnetworksanddeeplearning.com/chap5.html
[21] L. Datta, “A survey on activation functions and their relation with xavier and he
normal initialization,” ArXiv, vol. abs/2004.06632, 2020.
[22] “Colab documentation.” [Online]. Available: https://colab.research.google.com/
notebooks/intro.ipynb
[23] “歷史交通事故資料.” [Online]. Available: https://data.moi.gov.tw/MoiOD/Data/
DataDetail.aspx?oid=67781E29-8AAD-46A9-A2C8-C3F339592C27