臺灣博碩士論文加值系統

在快速演變的金融市場中，能夠準確預測金融時間序列中的短線交易信號對投資者來說至關重要。本論文提出了基於自適應局部漂移度密度同步漂移適應長短期記憶模型（adaptive LDD-DSDA LSTM），以應對金融時間序列數據中的概念漂移。該模型結合了自適應窗口算法（ADWIN）的連續漂移檢測能力和動態窗口調整功能，與基於局部漂移度的密度同步漂移適應（LDD-DSDA）技術以及在線更新閾值機制。實驗在真實世界的股票數據集上進行，包括聯發科技股份有限公司（MediaTek）、聯華電子股份有限公司、長榮海運股份有限公司、台灣積體電路製造股份有限公司（TSMC）、華邦電子股份有限公司和陽明海運股份有限公司。研究結果表明，在大多數情況下，本論文引入的模型的分類性能有所提升。

In the rapidly evolving landscape of financial markets, the ability to accurately predict swing trade signals in financial time series is important for investors. This thesis proposes the adaptive Local Drift Degree-based Density Synchronized Drift Adaptation Long Short-Term Memory (adaptive LDD-DSDA LSTM) model to tackle concept drift in financial time series data. This model integrates the continuous drift detection capabilities and dynamic window adjustment of the Adaptive Windowing (ADWIN) algorithm with the data density synchronization technique of Local Drift Degree-based Density Synchronized Drift Adaptation (LDD-DSDA) and online updating threshold mechanism. The experiments are simulated on real-world stock datasets, including MediaTek Inc. (MediaTek), United Microelectronics Corp., Evergreen Marine Corporation, Taiwan Semiconductor Manufacturing Company, Winbond Electronics Corp, and Yang Ming Marine Transport Corporation. The findings indicate that the classification performance of the model introduced in this paper has improved in most cases.

Content i

1. Introduction 1
1.1 Background 1
1.2 Motivation 4
1.3 Objective 5

2. Literature Review 7
2.1 Concept Drift Detection Algorithm 7
2.1.1 Error rate-based drift detection 7
2.1.2 Windows-based drift detection 10
2.1.3 Data distribution-based drift detection 12
2.1.4 Summary of concept drift detection algorithms 16
2.2 Prediction Model 18
2.2.1 Prediction model in drift detection algorithms 18
2.2.2 Stock prediction model 19

3. Methodology 22
3.1 Data preprocessing and pretraining model 22
3.1.1 Swing trade signal labeling 23
3.1.2 Feature selection and normalization 23
3.1.3 Pretraining model 24
3.2 Adaptive Windowing 26
3.2.1 Adaptive windowing (ADWIN) algorithm 26
3.2.2 Ensemble of univariate ADWIN detectors 27
3.3 Local Drift Degree-based Density Synchronized Drift Adaptation 29
3.3.1 Local Drift Degree (LDD) 29
3.3.2 LDD-based drifted instance selection algorithm (LDD-DIS) 29
3.3.3 LDD-based Density Synchronized Drift Adaptation (LDD-DSDA) 32
3.4 Adaptive Local Drift Degree-based Density Synchronized Drift Adaptation 33
3.4.1 Adaptive Local Drift Degree-based Density Synchronized Drift Adaptation 33
3.4.2 Online updating threshold mechanism 35

4. Metrics of Evaluation 36
4.1 Classification Metrics 36
4.2 Computation time 37

5. Experimentation 38
5.1 Datasets 38
5.2 Experimental setup 38
5.3 Comparative model 39
5.4 Experimental results and analysis 39

6. Conclusion 45

Reference 47

List of Figures
Fig. 1 Model accuracy decay along the time due to concept drift 2
Fig. 2 Types of concept drift. 3
Fig. 3 Landmark time window for drift detection. 8
Fig. 4 Two-time windows for concept drift detection. The recent window must be predefined. 11
Fig. 5 Two sliding window strategy to trigger data distribution-based algorithm while window length = 3. 13
Fig. 6 Visualization at depth 8 of the kdq-tree in [17]. 14
Fig. 7 An LSTM cell architecture 19
Fig. 8 The flow chart of pre-training model. 25
Fig. 9 Adaptive Windowing algorithm. 27
Fig. 10 An illustration of the ensemble ADWIN scheme. 28
Fig. 11 LDD-based drifted instance selection algorithm (LDD-DIS) algorithm 31
Fig. 12 LDD-based Density Synchronized Drift Adaptation (LDD-DSDA) algorithm 33
Fig. 13 Combining drift detection mechanism and LSTM in a data stream 34

List of Tables
TABLE 1 Summarization of concept drift detection algorithm 16
TABLE 2 Performance on MediaTek dataset 40
TABLE 3 Performance on UMC dataset 40
TABLE 4 Performance on Evergreen dataset 40
TABLE 5 Performance on TSMC dataset 41
TABLE 6 Performance on Winbond dataset 42
TABLE 7 Performance on Yang Ming dataset 42
TABLE 8 Average performance on all dataset 43

[1]S. Banik, N. Sharma, M. Mangla, S. N. Mohanty, and S. S., “LSTM based decision support system for swing trading in stock market,” Knowledge-Based Systems, vol. 239, p. 107994, Mar. 2022, doi: 10.1016/j.knosys.2021.107994.
[2]J. Shah, D. Vaidya, and M. Shah, “A comprehensive review on multiple hybrid deep learning approaches for stock prediction,” Intelligent Systems with Applications, vol. 16, p. 200111, Nov. 2022, doi: 10.1016/j.iswa.2022.200111.
[3]W. Jiang, “Applications of deep learning in stock market prediction: Recent progress,” Expert Systems with Applications, vol. 184, p. 115537, Dec. 2021, doi: 10.1016/j.eswa.2021.115537.
[4]D. P. Gandhmal and K. Kumar, “Systematic analysis and review of stock market prediction techniques,” Computer Science Review, vol. 34, p. 100190, Nov. 2019, doi: 10.1016/j.cosrev.2019.08.001.
[5]J. Lu, A. Liu, F. Dong, F. Gu, J. Gama, and G. Zhang, “Learning under Concept Drift: A Review,” IEEE Transactions on Knowledge and Data Engineering, vol. 31, no. 12, pp. 2346–2363, Feb. 2019, doi: 10.1109/TKDE.2018.2876857.
[6]X. You, M. Zhang, D. Ding, F. Feng, and Y. Huang, “Learning to Learn the Future: Modeling Concept Drifts in Time Series Prediction,” in Proceedings of the 30th ACM International Conference on Information & Knowledge Management, in CIKM ’21. New York, NY, USA: Association for Computing Machinery, Oct. 2021, pp. 2434–2443. doi: 10.1145/3459637.3482271.
[7]S. Agrahari and A. K. Singh, “Concept Drift Detection in Data Stream Mining : A literature review,” Journal of King Saud University - Computer and Information Sciences, vol. 34, no. 10, Part B, pp. 9523–9540, Nov. 2022, doi: 10.1016/j.jksuci.2021.11.006.
[8]A. Bifet and R. Gavaldà, “Learning from Time-Changing Data with Adaptive Windowing,” in Proceedings of the 2007 SIAM International Conference on Data Mining (SDM), 0 vols., in Proceedings. , Society for Industrial and Applied Mathematics, 2007, pp. 443–448. doi: 10.1137/1.9781611972771.42.
[9]A. Liu, Y. Song, G. Zhang, and J. Lu, “Regional Concept Drift Detection and Density Synchronized Drift Adaptation,” in Proceedings of the Twenty-Sixth International Joint Conference on Artificial Intelligence, Melbourne, Australia: International Joint Conferences on Artificial Intelligence Organization, Aug. 2017, pp. 2280–2286. doi: 10.24963/ijcai.2017/317.
[10]S. Hochreiter and J. Schmidhuber, “Long Short-Term Memory,” Neural Computation, vol. 9, no. 8, pp. 1735–1780, Jan. 1997, doi: 10.1162/neco.1997.9.8.1735.
[11]J. Gama, P. Medas, G. Castillo, and P. Rodrigues, “Learning with Drift Detection,” in Advances in Artificial Intelligence – SBIA 2004, A. L. C. Bazzan and S. Labidi, Eds., in Lecture Notes in Computer Science. Berlin, Heidelberg: Springer, 2004, pp. 286–295. doi: 10.1007/978-3-540-28645-5_29.
[12]M. Baena-García, J. Campo-Ávila, R. Fidalgo-Merino, A. Bifet, R. Gavald, and R. Morales-Bueno, “Early Drift Detection Method,” Jan. 2006.
[13]I. Frías-Blanco, J. del Campo-Ávila, G. Ramos-Jiménez, R. Morales-Bueno, A. Ortiz-Díaz, and Y. Caballero-Mota, “Online and Non-Parametric Drift Detection Methods Based on Hoeffding’s Bounds,” IEEE Transactions on Knowledge and Data Engineering, vol. 27, no. 3, pp. 810–823, Mar. 2015, doi: 10.1109/TKDE.2014.2345382.
[14]G. J. Ross, N. M. Adams, D. K. Tasoulis, and D. J. Hand, “Exponentially weighted moving average charts for detecting concept drift,” Pattern Recognition Letters, vol. 33, no. 2, pp. 191–198, Jan. 2012, doi: 10.1016/j.patrec.2011.08.019.
[15]R. C. Cavalcante and A. L. I. Oliveira, “An approach to handle concept drift in financial time series based on Extreme Learning Machines and explicit Drift Detection,” in 2015 International Joint Conference on Neural Networks (IJCNN), Jul. 2015, pp. 1–8. doi: 10.1109/IJCNN.2015.7280721.
[16]K. Nishida and K. Yamauchi, “Detecting Concept Drift Using Statistical Testing,” in Discovery Science, V. Corruble, M. Takeda, and E. Suzuki, Eds., in Lecture Notes in Computer Science. Berlin, Heidelberg: Springer, 2007, pp. 264–269. doi: 10.1007/978-3-540-75488-6_27.
[17]T. Dasu, S. Krishnan, S. Venkatasubramanian, and K. Yi, “An Information-Theoretic Approach to Detecting Changes in MultiDimensional Data Streams,” Interfaces, Jan. 2006.
[18]D. D. H. Bailey, “Advances in Financial Machine Learning”.
[19]W. J. Faithfull, J. J. Rodríguez, and L. I. Kuncheva, “Combining univariate approaches for ensemble change detection in multivariate data,” Information Fusion, vol. 45, pp. 202–214, Jan. 2019, doi: 10.1016/j.inffus.2018.02.003.