Performance Comparison of Deep-Learning Algorithms for Groundnut Price Forecasting: Evidence from Rajkot and Junagadh Markets
9 / 15
Keywords:
AGMARKNET; Deep learning; Groundnut; Price forecasting; Sensitivity analysis; Stacked LSTM; Time-series modelling.Abstract
Reliable price forecasts can improve marketing, storage and policy decisions in oilseed value chains. This study benchmarks five deep-learning
architectures—deep neural network (DNN), recurrent neural network (RNN), long short-term memory (LSTM), gated recurrent unit (GRU) and
Stacked LSTM—using daily groundnut modal prices from Rajkot and Junagadh markets. Each market contains 7,402 calendar-daily observations
from 1 January 2004 to 6 April 2024. A reconstruction audit identified 801 Rajkot and 1,757 Junagadh prices absent from the surviving incomplete
copy; the verified complete companion series was retained, and the paper-consistent causal trailing five-day smoother was preserved. Models were
evaluated on an identical chronological 85:15 train–test split using RMSE, MAPE, sMAPE and mean directional accuracy. Stacked LSTM gave
the lowest RMSE (180 Rajkot; 185 Junagadh), lowest MAPE (5%; 7%) and highest directional accuracy (88%; 89%). Alternative gap-filling rules
changed market means by less than 0.75% and sample standard deviations by less than 0.83%, supporting descriptive robustness. The contribution
is a like-for-like, two-market comparison over a common 20-year daily horizon, with both magnitude and direction metrics and a transparent
reconstruction sensitivity check.
Downloads
References
1. AGMARKNET. (2024). Agriculture Marketing Information Network.
2. Directorate of Marketing and Inspection, Government of India.
4. Cho, K., van Merriënboer, B., Gulcehre, C., Bahdanau, D., Bougares,
5. F., Schwenk, H., and Bengio, Y. (2014). Learning phrase
6. representations using RNN encoder–decoder for statistical
7. machine translation. In Proceedings of the 2014 Conference on
8. Empirical Methods in Natural Language Processing (EMNLP)
9. (pp. 1724–1734). Association for Computational Linguistics.
10. https://doi.org/10.3115/v1/D14-1179
11. Das, P., Jha, G.K. and Lama, A. (2023). Empirical mode decomposition
12. based hybrid machine learning models for agricultural commodity
13. price forecasting. Statistics and Applications, 21(1), 99-112.
14. Goodfellow, I., Bengio, Y., and Courville, A. (2016). Deep learning.
15. MIT Press.
16. Graves, A., Mohamed, A.-R., and Hinton, G. (2013). Speech recognition
17. with deep recurrent neural networks. In 2013 IEEE International
18. Conference on Acoustics, Speech and Signal Processing (pp. 6645
19. 6649). IEEE. https://doi.org/10.1109/ICASSP.2013.6638947
20. Harshith, N., and Kumari, P. (2024). Memory based neural network
21. for cumin price forecasting in Gujarat. Journal of Agriculture
22. and Food Research, 15, Article 101020. https://doi.org/10.1016/j.
23. jafr.2024.101020
24. Hochreiter, S., and Schmidhuber, J. (1997). Long short-term memory.
25. Neural Computation, 9(8), 1735–1780. https://doi.org/10.1162/
26. neco.1997.9.8.1735
27. Jha, G.K., and Sinha, K. (2014). Time-delay neural networks for time
28. series prediction: An application to the monthly wholesale price
29. of oilseeds in India. Neural Computing and Applications, 24(3-4),
30. 563-571. https://doi.org/10.1007/s00521-012-1264-z
31. Kingma, D.P., and Ba, J. (2015). Adam: A method for stochastic
32. optimization. In Proceedings of the 3rd International Conference .
33. Kurumatani, K. (2020). Time-series forecasting of agricultural product
34. prices based on recurrent neural networks and its evaluation
35. method. SN Applied Sciences, 2(8), Article 1434. https://doi.
36. org/10.1007/s42452-020-03225-9
37. LeCun, Y., Bengio, Y., and Hinton, G. (2015). Deep learning. Nature,
38. 521(7553), 436-444. https://doi.org/10.1038/nature14539
39. Muthuswamy, K., Dolli, S., Khandeparkar, K., Gowda, C., Narasimma
40. Bharathi, V.P., Shivakumar, K.M., Sumathi, C. S., Appavu,
41. S., Natarajan, B., and Sivakumar, K. (2025). Weather-driven
42. groundnut price forecasting and profitability assessment of
43. cropping patterns in Tamil Nadu using boosting algorithms.
44. Scientific Reports, 15(1), Article 33880. https://doi.org/10.1038/
45. s41598-025-08573-3
46. Nair, V., and Hinton, G.E. (2010). Rectified linear units improve
47. restricted Boltzmann machines. In Proceedings of the 27th
48. International Conference on Machine Learning (ICML-10)
49. (pp. 807-814). Omnipress.
50. Nayak, G.H.H., Alam, M.W., Singh, K.N., Avinash, G., Kumar, R.R.,
51. Ray, M., and Deb, C. K. (2024). Exogenous variable driven deep
52. learning models for improved price forecasting of TOP crops
53. in India. Scientific Reports, 14(1), Article 17203. https://doi.
54. org/10.1038/s41598-024-68040-3
55. NN-SVG. (n.d.). Publication-ready neural-network architecture
56. schematics. https://alexlenail.me/NN-SVG/
57. Paul, R.K., Yeasin, M., Kumar, P., Kumar, P., Balasubramanian, M.,
58. Roy, H.S., Paul, A.K., and Gupta, A. (2022). Machine learning
59. techniques for forecasting agricultural prices: A case of brinjal in
60. Odisha, India. PLOS ONE, 17(7), Article e0270553. https://doi.
61. org/10.1371/journal.pone.0270553
62. Rumelhart, D.E., Hinton, G.E., and Williams, R.J. (1986). Learning
63. representations by back-propagating errors. Nature, 323(6088),
64. 533-536. https://doi.org/10.1038/323533a0
65. Werbos, P.J. (1990). Backpropagation through time: What it does and
66. how to do it. Proceedings of the IEEE, 78(10), 1550-1560. https://
67. doi.org/10.1109/5.58337