State-Space Forecasting for Nonstationary Multivariate Time Series with Adaptive Masking

Authors

  • Lei Lu Division of AI, School of Data Science, Lingnan University, Hong Kong
  • Michelle Wong Division of AI, School of Data Science, Lingnan University, Hong Kong

DOI:

https://doi.org/10.66421/int.j.datasci.intell.syst47

Keywords:

State-Space Models, Multivariate Time Series, Nonstationary Processes, Adaptive Masking, Predictive Modeling

Abstract

Multivariate time series forecasting remains a critical yet highly challenging domain within machine learning and statistical modeling, primarily due to the ubiquitous presence of nonstationarity and complex inter-variable dependencies. Real-world temporal data collected from financial markets, meteorological sensors, and large-scale infrastructure networks frequently exhibit distribution shifts, regime changes, and structural breaks that degrade the predictive accuracy of conventional models. While deep state-space models have recently demonstrated exceptional capabilities in capturing long-range dependencies with linear computational complexity, their application to highly nonstationary environments often results in suboptimal performance due to their inherent assumption of stable transition dynamics. To address these limitations, this paper proposes a novel framework that integrates an adaptive masking mechanism directly into the state-space transition formulation. The proposed methodology dynamically learns to mask irrelevant or highly volatile observation components, thereby preventing transient noise and sudden distributional shifts from corrupting the latent state representations. Through extensive theoretical analysis and empirical evaluation across multiple benchmark datasets, the proposed adaptive masking state-space architecture is shown to significantly outperform existing state-of-the-art architectures, including canonical recurrent networks, advanced attention-based transformers, and standard state-space models. The rigorous experimental results confirm that dynamically modulating the observational input based on localized stationarity metrics dramatically enhances the robustness and precision of long-term forecasts.

References

1. Liang, R., Xu, S., Xie, C., Chen, J., Ren, F., Yang, S., & Yabe, T. (2026). Abstain Mask Retain Core: Time Series Prediction by Adaptive Masking Loss with Representation Consistency. Advances in Neural Information Processing Systems, 38, 99445-99471.

2. Jin, X., Diao, L., Xiao, Q., Hu, Y., Zhang, Z., Liu, Y., & Gu, H. (2025). CCAD: Compressed Global Feature Conditioned Anomaly Detection. arXiv preprint arXiv:2512.21459.

3. Xiaoteng Ma, Yiqin Yang, Hao Hu, Qihan Liu, Jun Yang, Chongjie Zhang, Qianchuan Zhao, and Bin Liang. Offline reinforcement learning with value-based episodic memory. arXiv preprint arXiv:2110.09796, 2021b.

4. Chen, L., Ge, L., Jiang, X., Li, H., & Haardt, M. (2025). Slim is better: Transform-based tensor robust principal component analysis. IEEE Transactions on Signal Processing.

5. Qu, D., Ma, Y., Yan, J., & Pyrozhenko, M. (2026). TriMeta-BFNet: A Tri-Meta Stacked Atypical-Frequency Bayesian Fourier Neural Network for Hallucination-Resistant Community Detection. Mathematics, 14 (13), 2283.

6. Wang, J., Chen, L., & Al Faruque, M. A. (2023, October). Domino: Domain-invariant hyperdimensional classification for multi-sensor time series data. In2023 IEEE/ACM International Conference on Computer Aided Design (ICCAD)(pp. 1-9). IEEE.

7. Gu, C., Zhang, W., Huang, Z., Kou, J., Liu, Z., Zhao, C., Liu, C., Zhang, L., Lin, W., Wang, Z., Deng, J., Xie, Y., Huang, G., Zhang, C., Lu, X., Wang, C., Zhang, Z., Yuan, H., Duan, X., & Fang, Y. (2024). LENS: Layers of evaluation of hallucination in GenAI systems. In 2024 7th International Conference on Universal Village (UV) (pp. 1–85). IEEE. https://doi.org/10.1109/UV63228.2024.11189150

8. Kingma, D. P., & Ba, J. (2015). Adam: A method for stochastic optimization. In Proceedings of ICLR.

9. Tibshirani, R. (1996). Regression shrinkage and selection via the lasso. Journal of the Royal Statistical Society: Series B, 58(1), 267–288.

10. Zhu, R., Ma, Z., Wu, J., Gao, J., Wang, J., Lin, D., & He, C. (2025, April). Utilize the flow before stepping into the same river twice: Certainty represented knowledge flow for refusal-aware instruction tuning. In Proceedings of the AAAI Conference on Artificial Intelligence (Vol. 39, No. 24, pp. 26157-26165).

11. Ma, Y., & Qu, D. (2026, February). IMFM-GNN: Interpretable Multimodel GNN Via Fourier-Markov Dynamics for Community Detection in Sustainable Energy Networks. In 2026 International Conference on Artificial Intelligence, Computer, Data Sciences and Applications (ACDSA) (pp. 1-6). IEEE.

12. Luo H, Li Q, Zhou H, et al. Weak single phase-to-ground fault time detection with learnable-parameter-driven DSP-enhanced transformer[J]. IEEE Transactions on Smart Grid, 2025.

13. Pei, Zhengqi, Qingming Huang, and Shuhui Wang. "When LLMs develop languages: Symbolic communication for efficient multi-agent reasoning." arXiv preprint arXiv:2606.29354 (2026).

14. Han, Z., Chen, W., Han, Y., Mao, R., & Qin, J. (2026). Fast Diversified Top-k Rule Discovery via User-Guided Embeddings. IEEE Transactions on Knowledge and Data Engineering.

15. Yang Y, Hu H, Mao Y, et al. OPRIDE: Offline Preference-based Reinforcement Learning via In-Dataset Exploration[J]. arXiv preprint arXiv:2604.02349, 2026.

16. Zhang, J., Shi, Y., Ma, Y., Xu, L., Yu, J., & Wang, J. (2023, June). Ikol: Inverse kinematics optimization layer for 3d human pose and shape estimation via gauss-newton differentiation. In Proceedings of the AAAI conference on artificial intelligence (Vol. 37, No. 3, pp. 3454-3462).

17. Wang, J., & Faruque, M. A. A. (2023). Robust and scalable hyperdimensional computing with brain-like neural adaptations.arXiv preprint arXiv:2311.07705.

18. Hu, H., Yang, Y., Zhao, Q., & Zhang, C. (2022, June). On the role of discount factor in offline reinforcement learning. In International conference on machine learning (pp. 9072-9098). PMLR.

19. Yang, Y., Ma, X., Li, C., Zheng, Z., Zhang, Q., Huang, G., ... & Zhao, Q. (2021). Believe what you see: Implicit constraint approach for offline multi-agent reinforcement learning. Advances in Neural Information Processing Systems, 34, 10299-10312.

20. Wei, W. D., Yang, X. W., Shao, J. J., & Guo, L. Z. (2025, August). Curriculum abductive learning for mitigating reasoning shortcuts. In Proceedings of the Thirty-Fourth International Joint Conference on Artificial Intelligence, IJCAI-25 (pp. 6533-6541).

21. Cao, C., Xue, X., Feng, L., & Tan, K. C. (2025, October). Why knowledge transfer works in oracle-based optimization: A brief analysis. In2025 International Conference on Machine Intelligence and Nature-Inspired Computing (MIND)(pp. 290-291). IEEE.

22. Xue, X., Hu, Y., Feng, L., Zhang, K., Song, L., & Tan, K. C. (2024). Surrogate-assisted search with competitive knowledge transfer for expensive optimization. IEEE Transactions on Evolutionary Computation.

23. Zhang, Y., Zhang, Y., Zhao, M., Feng, S., & Cheung, Y. M. (2026). How to Ask the AI: A User Perspective Survey for Large Language Model Prompting.IEEE Transactions on Artificial Intelligence.

24. Han, Z., Chen, W., & Han, Y. (2024, August). Data Quality-Driven Top-k Rule Discovery. In International Conference on Machine Learning, Cloud Computing and Intelligent Mining (pp. 24-38). Singapore: Springer Nature Singapore.

25. Xu, Yuanchao, et al. "Reskoopnet: Learning koopman representations for complex dynamics with spectral residuals." arXiv preprint arXiv:2501.00701 (2025).

26. Liu, Chenchen, Fuqiang Liu, and Hai Li. "Beyond CMOS: memristor and its application for next generation storage and computing." Electrochemical Society Transactions 233 85, no. 6 (2018): 115-125.

27. Qu, D., Ma, Y., & Wang, Y. (2026). STR-GNN: stability-regularized graph neural networks for suppressing spurious temporal variations in dynamic community detection. Scientific Reports.

28. Wang, J., & Al Faruque, M. A. (2025, April). Transformer-based contrastive meta-learning for low-resource generalizable activity recognition. In ICASSP 2025-2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)(pp. 1-5). IEEE.

29. Ma, Y., Qu, D., & Wang, Y. (2026). HALO-GNN: hallucination-resistant temporal graph neural networks for dynamic community detection.Scientific Reports.

Downloads

Published

2026-08-05