International Journal of

ADVANCED AND APPLIED SCIENCES

EISSN: 2313-3724, Print ISSN: 2313-626X

Frequency: 12

line decor
  
line decor

 Volume 13, Issue 10 (October 2026), Pages: 20-32

----------------------------------------------

 Original Research Paper

An adjoint sensitivity-based Lévy-induced stochastic differential equation network for agricultural price prediction

 Author(s): 

Nur Amalina Shafie 1, *, Mohd Tahir Ismail 2, Isnewati Ab Malek 1, Zahari Md Rodzi 1, 3

 Affiliation(s):

1Faculty of Computer and Mathematical Sciences, Universiti Teknologi MARA, Seremban, Malaysia
2School of Mathematical Sciences, Universiti Sains Malaysia, Penang, Malaysia
3Accounting Research Institute (ARI), Universiti Teknologi MARA, Shah Alam, Malaysia

 Full text

    Full Text - PDF

 * Corresponding Author. 

   Corresponding author's ORCID profile:  https://orcid.org/0000-0003-4522-1758

 Digital Object Identifier (DOI)

 
https://doi.org/10.21833/ijaas.2026.10.003

 Abstract

Agricultural commodity prices exhibit non-linearity, heavy-tailed return distributions, and occasional large jumps that standard Gaussian models cannot adequately capture. Existing stochastic neural networks, including the Lévy-Induced Stochastic Differential Equation Network (LDE-Net), address these characteristics but suffer from gradient instability during backpropagation through stochastic Lévy trajectories. This study proposes the Adjoint LDE-Net, extending the adjoint sensitivity method from deterministic Neural ODEs to the Lévy stochastic setting. Its backward adjoint SDE computes parameter gradients without traversing the forward Lévy trajectory, reducing memory complexity from O(N) to O(1) and improving gradient stability. Two datasets are evaluated using a 70:30 train-test split: the IMF Food and Beverage Price Index (434 monthly observations) and the FAMA Malaysian Tomato Price series (208 weekly observations). Both series exhibit chaos according to the Wolf method and stationarity in log-returns according to the Augmented Dickey-Fuller test. Finite-difference sensitivity analysis yields ∂f/∂σ = ∂f/∂μ = 1.000 for both datasets, indicating the importance of the diffusion pathway. Neural SDE achieves the best predictive performance on both datasets, while Adjoint LDE-Net ranks second on the IMF dataset. Both stochastic neural models outperform ARIMA and GBM by more than an order of magnitude in MSE on the IMF dataset. The proposed method provides a theoretically grounded, memory-efficient approach to Lévy SDE training. However, its predictive accuracy does not exceed that of Neural SDE, and instability on the Tomato series highlights the need for α calibration and gradient clipping.

 © 2026 The Authors. Published by IASE.

 This is an open access article under the CC BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0/).

 Keywords

Adjoint sensitivity, Lévy processes, Stochastic differential equations, Neural networks, Agricultural price forecasting

 Article history

Received 23 May 2026, Received in revised form 17 September 2026, Accepted 10 October 2026

 Acknowledgment

The authors gratefully acknowledge the support of the Fundamental Research Grant Scheme for Early Career Researchers (FRGS-EC), Ministry of Higher Education Malaysia (Reference Code: FRGS-EC/1/2024/STG06/UITM/02/29). The authors thank the Federal Agricultural Marketing Authority (FAMA) for providing the Tomato Price dataset. 

 Compliance with ethical standards

 Conflict of interest: The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

 Citation:

Shafie NA, Ismail MT, Malek IA, and Rodzi ZM (2026). An adjoint sensitivity-based Lévy-induced stochastic differential equation network for agricultural price prediction. International Journal of Advanced and Applied Sciences, 13(10): 20-32

  Permanent Link to this page

---------------------------------------------- 

 References (18)

  1. Bhagat PR, Naz F, and Magda R (2022). Artificial intelligence solutions enabling sustainable agriculture: A bibliometric analysis. PLOS ONE, 17(6): e0268989. https://doi.org/10.1371/journal.pone.0268989   [Google Scholar] PMid:35679287 PMCid:PMC9182339
  2. Box GEP, Jenkins GM, Reinsel GC, and Ljung GM (2016). Time series analysis: Forecasting and control. 5th Edition, John Wiley & Sons, Inc., Hoboken, USA.   [Google Scholar]
  3. Cao L (1997). Practical method for determining the minimum embedding dimension of a scalar time series. Physica D: Nonlinear Phenomena, 110(1-2): 43-50. https://doi.org/10.1016/S0167-2789(97)00118-8   [Google Scholar]
  4. Chen RTQ, Rubanova Y, Bettencourt J, and Duvenaud DK (2018). Neural ordinary differential equations. In the Proceedings of the 32nd Conference on Neural Information Processing Systems (NeurIPS 2018), Montréal, Canada, 31: 6571–6583.   [Google Scholar]
  5. Duan J (2015). An introduction to stochastic dynamics. Cambridge University Press, New York, USA.  [Google Scholar]
  6. E W, Ma C, and Wu L (2022). The Barron space and the flow-induced function spaces for neural network models. Constructive Approximation, 55: 369–406. https://doi.org/10.1007/s00365-021-09549-y   [Google Scholar]
  7. Heston SL (1993). A closed-form solution for options with stochastic volatility with applications to bond and currency options. The Review of Financial Studies, 6(2): 327-343. https://doi.org/10.1093/rfs/6.2.327   [Google Scholar]
  8. Jia J and Benson AR (2019). Neural jump stochastic differential equations. In the Proceedings of the 33rd Conference on Neural Information Processing Systems (NeurIPS 2019), Vancouver, Canada, 32: 9843–9854.   [Google Scholar]
  9. Kong L, Sun J, and Zhang C (2020). SDE-Net: Equipping deep neural networks with uncertainty estimates. ArXiv Preprint ArXiv:2008.10546. https://doi.org/10.48550/arXiv.2008.10546   [Google Scholar]
  10. Liu W (2019). Strong convergence rate of Euler-Maruyama method for stochastic differential equations with Hölder continuous drift coefficient driven by symmetric α-stable process. ArXiv Preprint ArXiv:1901.08742. https://doi.org/10.48550/arXiv.1901.08742   [Google Scholar]
  11. Mao L, Huang Y, Zhang X, Li S, and Huang X (2022). ARIMA model forecasting analysis of the prices of multiple vegetables under the impact of the COVID-19. PLOS ONE, 17(7): e0271594. https://doi.org/10.1371/journal.pone.0271594   [Google Scholar] PMid:35901077
  12. Merton RC (1976). Option pricing when underlying stock returns are discontinuous. Journal of Financial Economics, 3(1-2): 125-144. https://doi.org/10.1016/0304-405X(76)90022-2  [Google Scholar]
  13. Paul RK, Yeasin M, Kumar P, Kumar P, Balasubramanian M, Roy HS, Paul AK, and Gupta A (2022). Machine learning techniques for forecasting agricultural prices: A case of brinjal in Odisha, India. PLOS ONE, 17(7): e0270553. https://doi.org/10.1371/journal.pone.0270553   [Google Scholar] PMid:35793366 PMCid:PMC9258887
  14. Qazi S, Khawaja BA, and Farooq QU (2022). IoT-equipped and AI-enabled next generation smart agriculture: A critical review, current challenges and future trends. IEEE Access, 10: 21219-21235. https://doi.org/10.1109/ACCESS.2022.3152544   [Google Scholar]
  15. Samorodnitsky G and Taqqu MS (1994). Stable non-Gaussian random processes: Stochastic models with infinite variance. CRC Press, Boca Raton, USA.   [Google Scholar]
  16. Wang JZ, Wang JJ, Zhang ZG, and Guo SP (2011). Forecasting stock indices with back propagation neural network. Expert Systems with Applications, 38(11): 14346-14355. https://doi.org/10.1016/j.eswa.2011.04.222   [Google Scholar]
  17. Wolf A, Swift JB, Swinney HL, and Vastano JA (1985). Determining Lyapunov exponents from a time series. Physica D: Nonlinear Phenomena, 16(3): 285-317. https://doi.org/10.1016/0167-2789(85)90011-9   [Google Scholar]
  18. Yang L, Gao T, Lu Y, Duan J, and Liu T (2023). Neural network stochastic differential equation models with applications to financial data forecasting. Applied Mathematical Modelling, 115: 279-299. https://doi.org/10.1016/j.apm.2022.11.001   [Google Scholar]