S.T. Tsaplin1, D.V. Berezkin2, G.P. Mojarov3

1-3 Bauman Moscow State Technical University (Moscow, Russia)

1 tsaplin@bmstu.ru, 2 berezkind@bmstu.ru, 3 mojarov@bmstu.ru

This paper presents a comprehensive comparative analysis of existing methods for augmenting short time series, addressing the critical challenge of forecasting socio-economic indicators with limited historical data. The study systematically reviews and evaluates various augmentation techniques ranging from classical statistical methods to advanced neural network approaches, specifically focusing on their applicability to extremely short time series with fewer than 40 data points. The analyzed methods include traditional techniques such as noising (adding random noise to existing data), scaling (changing amplitude while preserving shape), cropping (signal truncation), permutation (temporal shuffling), rotation (data flip ping), and bootstrap approaches (statistical resampling). Additionally, the analysis covers sophisticated generative models including TSA -GAN (with monitor and restorer components), T -CGAN (conditional generative model), and TimeGAN (combining supervised a nd unsupervised learning). The research reveals significant limitations in current augmentation approaches. While sophisticated GAN -based models can achieve forecast accuracy up to 0.97 when sufficient data points (19 –40) are available, they fail to effect ively address the augmentation needs for extremely short time series. Traditional methods, though computationally efficient, offer limited capabilities for generating meaningful synthetic data that preserves underlying temporal patterns and dependencies. The study identifies specific applicability domains for each method: classical methods are suitable for analytical tasks like trend detection and addressing class imbalance, while advanced neural approaches excel at generating similar time series when minim um data requirements are met. The research highlights a critical gap in the literature regarding augmentation methods specifically de signed for extremely short time series, particularly relevant in educational planning where institutions must make strategic decisions based on limited historical enrollment data. The paper concludes by proposing future research directions involving combination strategies that integrate simpler and more complex augmentation methods to improve forecasting performance for short time series applications. This research provides valuable insights for practitioners working with limited historical data in various socio-economic forecasting applications.

Tsaplin S.T., Berezkin D.V., Mozharov G.P. Comparative analysis of short time series augmentation methods. Dynamics of complex systems. 2026. V. 20. № 2. P. 70−78. DOI: 10.18127/j19997493-202602-08 (in Russian).

1. Qingsong Wen, Liang Sun, Fan Yang, Xiaomin Song, Jingkun Gao, Xue Wang, Huan Xu. Time Series Data Augmentation for Deep Learning: A Survey. Thirtieth International Joint Conference on Artificial Intelligence (IJCAI-21). 2020. P. 4653–4660.
2. Annaki I., Rahmoune M., Bourhaleb M. Overview of Data Augmentation Techniques in Time Series Analysis. International Journal of Advanced Computer Science and Applications. 2024. V. 15. P. 10123–10145.
3. Ana Lazcano de Rojas. Data augmentation in economictime series: Behavior and improvements in predictions. AIMS Mathematics. 2023. V. 8. P. 24528–24544.
4. Shiyu Liu, Hongyan Qiao, Lianhong Yuan, Yuan Yuan, Jun Liu. Research on dataaugmentation algorithm for timeseries based on deeplearning. KSII Transactions on Internet and Information Systems. 2023. V. 17. P. 1530–1544.
5. Monitoring kachestva priema v VUZy`. URL: https://ege.hse.ru/rating/2022/91645021/all/
6. Flores A., Tito-Chura H., Apaza-Alanoca H. Data Augmentation for Short-Term Time Series Prediction with Deep Learning. In: Arai K. (eds) Intelligent Computing. Lecture Notes in Networks and Systems. 2021. V. 284.
7. Rashid K.M., Louis J. Window-warping: atimeseries dataaugmentation of imu data for construction equipment activity identification. 36th International Symposium on Automation and Robotics in Construction (ISARC). 2019. V. 36. P. 651–657.
8. Xue Ying. An Overview of Overfitting and its Solutions. Journal of Physics Conference Series. 2019.
9. Le Guennec A., Malinowski S., Tavenard R. Data Augmentation for Time Series Classification using C onvolutional Neural Networks. ECML/PKDD Workshop on Advanced Analytics and Learning on Temporal Data. 2016. 1
10. Zhicheng Cui, Wenlin Chen. Multiscale convolutional neural networks for timeseries classification. 2016. URL: https://arxiv.org/abs/1603.06995 1
11. Ekin D Cubuk, Barret Zoph, Jonathon Shlens, and Quoc V Le. RandAugment: Practical automated data augmentation with areducedsearch space. In 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshop. 2020. P. 3008–3017. 1
12. Jingkun Gao, Xiaomin Song, Qingsong Wen, Pichao Wang, Liang Sun, and Huan Xu. Robusttad: Robust timeseries anomalydetection via decomposition and convolutional neural networks. 6th KDD MileTS Workshop on Mining and Learning from Time Series. 2020. P. 1–6. 1
13. Um T.T., Pfister F.M., Pichler D., Endo S., Lang M., Hirche S., Fietzek U., Kuli´c D. Data augmentation of wearablesensor data for parkinson’sdiseasemonitoring using convolutional neural networks. Proceedings of the 19th ACM International Conference on M ultimodal Interaction. 2017. P. 216– 2 2
14. Comparative analysis of methods of augmentation of shorttime series Dynamics of complex systems / Dinamika slozhnykh sistem, V. 20, № 2, 2026, p. 70−78 78 1
15. Fawaz H.I., Forestier G., Weber J., Idoumghar L., Muller P. -A. Data augmentation using synthetic data for timeseries classification with deepresidual networks. 2018. URL: https://arxiv.org/abs/1808.02455 1
16. Iwana B.K., Uchida S. An empirical survey of dataaugmentation for timeseries classification with neural networks. PLoS ONE 16(7). 2021. 1
17. Iwana B.K., Uchida S. Time seriesdata augmentation for neural networks by timewarping with adiscriminative teacher. ICPR. 2021. P. 3558–3565. 1
18. Rashid K.M., Louis J. Time-warping: A timeseries dataaugmentation of IMU data for construction equipment activity identification. 36th International Symposium on Automation and Robotics in Construction. 2019. 1
19. Athanasopoulos G., Song H., Sun J.A. Bagging in tourismdemand modeling and forecasting. Journal of Travel Research 57(1). 2017. P. 52–68. 1
20. Bergmeir C., Hyndman R.J., Ben´ıtez J.M. Bagging exponential smoothing methods using STL decomposition and Box –Cox transformation. International Journal of Forecasting 32(2). 2016. P. 303–312. 2
21. Dongsheng Luo, Wei Cheng, Yingheng Wang, Dongkuan Xu, Jingchao Ni, Wenchao Yu, Xuchao Zhang, Yanchi Liu, Yuncong Chen, Haifeng Chen, Xiang Zhang. Time Series Contrastive Learning with Information-Aware Augmentations. 2023. URL: https://arxiv.org/abs/2303.11911 2
22. Shivam Grover, Amin Jalali, Ali Etemad. Segment, Shuffle, and Stitch. A Simple Mechanism for Improving Time-Series Representations. 2024. URL: https://arxiv.org/html/2405.20082v1 2
23. Uchitomi Hirotaka & Ming Xianwen & Zhao Changyu & Ogata Taiki & Miyake Yoshihiro. Classification of mild Parkinson’sdisease: dataaugmentation of time-seriesgait dataobtained via inertialmeasurement units. Scientific Reports. 13. 2023. 2
24. Iftikhar N., Liu X., Danalachi S., Nordbjerg F.E., Vollesen J.H. A scalablesmart meterdata generator using spark. On the Move to Meaningful Internet Systems. OTM Conferences, Springer International Publishing. 2017. P. 21–36. 2
25. Cao P., Li X., Mao K., Lu F., Ning G., Fang L., Pan Q. A noveldata augmentation method to enhancedeep neural networks for detection of atrialfibrillation. Biomedical Signal Processing and Control 56. 2020. 2
26. Maer A.V., Simaxin V.A. Neparametricheskie datchikidlya sluchajny`xstacionarny`xprocessov. Sibirskij ae`rokosmicheskij zhurnal. 2010. S. 46–49. 2
27. Kim M. & Jeong C.Y. Label-preserving data augmentation for mobilesensor data. Multidimens. Systems and Signal Processing. 2021. 32. P. 115–129. 2
28. Terent`eva E.S. Primenenie butstrep-metoda vneparametricheskom modelirovanii sistem pri nalic hii propuskov danny`x. Reshetnevskie chteniya. 2010. S. 430–432. 2
29. Shuang Wu, Chi-Hua Wang, Yuantong Li, Guang Cheng. Residual Bootstrap Exploration for Stochastic Linear Bandit. 2020. URL: https://arxiv.org/pdf/2202.11474 2
30. Ty`chkov A.Yu. Primenenie modificirovannogo preobrazovaniya Gil`berta–Xuanga dlyaresheniya zadachcifrovoj obrabotki medicinskix signalov. Izvestiya vy`sshixuchebny`xzavedenij. Povolzhskij region. 2018. S. 70–80. 3
31. Kan Sh.Ch., Mikulovich A.V., Mikulovich V.I. Analiz nestacionarny`xsignalo v na osnove preobrazovaniya Gil`berta –Xuanga. Informatika. 2010. № 2. S. 25–35. 3
32. Pan Q., Li X., Fang L. Data augmentation for deeplearning-based ecg analysis. Feature Engineering and Computational Intelligence in ECG Monitoring. 2020. P. 91–111. 3
33. Rosario Ryan. A Data Augmentation Approach to Short Text Classification. 2017. URL: https://www.researchgate.net/publication/ 316714414_A_Data_Augmentation_Approach_to_Short_Text_Classification 3
34. Norden E Huang, Zheng Shen, Steven R Long, Manli C Wu, Hsing H Shih, Qua nan Zheng, Nai-Chyuan Yen, Chi Chao Tung, and Henry H Liu. The empirical mode decomposition and the hilbertspectrum for nonlinear and non-stationary time series analysis. Proceedings of the Royal Society of London. Series A: mathematical, physical and engineering sciences. 1998. P. 903–995. 3
35. Safiullin N.T., Porshnev S.V., Kleeorin N. Data analysis of sunspottime series with SSA and HHT information adaptive method. 2017. URL: https://ceur-ws.org/Vol-2005/paper-13.pdf 3
36. Norden Huang. Applications of Hilbert-Huang transform to non-stationary financial timeseries analysis. Applied Stochastic Models in Business and Industry, 19. 2003. P. 245–268. 3
37. Kurbatsky V.G., Sidorov D.N., Spiryaev V.A., Tomin N. Forecasting Nonstationary Time Series Based on Hilbert-Huang Transform and Machine Learning. Automation and Remote Control 75(5). 2014. P. 922–934. 3
38. Goodfellow Ian & Pouget-Abadie Jean & Mirza Mehdi & Xu Bing & Warde-Farley David & Ozair Sherjil & Courville Aaron & Bengio Y. Generative Adversarial Networks. Advances in Neural Information Processing Systems. 3. 2014. 3
39. Ramponi G., Protopapas P., Brambilla M., Janssen R. T-CGAN: Conditional Generative Adversarial Network for Data Augmentation in Noisy Time Series with Irregular Sampling. 2019. URL: https://arxiv.org/pdf/1811.08295 3
40. Li Z., Ma C., Shi X., Zhang D., Li W. and Wu L. TSA-GAN: A Robust Generative Adversarial Networks for Time Series Augmentation. 2021 International Joint Conference on Neural Networks (IJCNN). 2021. P. 1–8. 4
41. Jinsung Yoon, Daniel Jarrett. Time-series Generative Adversarial Networks. Neural Information Processing Systems (NeurIPS). 2019. 4
42. Laftchiev E., Liu Y. Finding Multidimensional Patterns in Multidimensional Time Series. KDD Workshop on Mining and Learning From Time Series. 2018.