Forecasting Soybean Production in Turkey: A Comparative Analysis of Automated and Traditional Methods
This is the English version of the README file.
For the Turkish version, click here.
The full text of the article related to this study can be accessed via the following link: https://dergipark.org.tr/en/pub/iutbd/issue/85235/1494324
Please cite this study if you use it in your research as follows:
Duman, H. (2024). Forecasting Soybean Production in Turkey: A Comparative Analysis of Automated and Traditional Methods. Agro Science Journal of Igdir University, 2(1), 90-102.
Turkey’s climate and soil are well-suited for oilseed crops, which are vital for various industries and human and animal diets. Soybeans, a legume, hold particular nutritional value among oilseeds. While existing research covers soybean production in Turkey, this study aims to: a) evaluate production levels using different forecasting algorithms to identify the most accurate model, and b) based on the chosen model, forecast future production and assess the current and future entrepreneurial potential of the soybean industry in Turkey. Soybean production data (1990-2022) from TURKSTAT was divided into training (n=26) and test (n=8) sets for cross-validation. By applying univariate time series methods, including ARIMA, SES, NNAR, MN, and Naive to the training dataset, it was found that ARIMA(1,1,1) performed best according to test set RMSE values. The performance ranking (in terms of RMSE) was as follows: ARIMA (13019) < SES (13888) < Naive (14240) < NNAR (58393) < MN (80418). Notably, for this dataset, the performance of automated processes was relatively worse than that of manual methods, suggesting that relying solely on automated methods may lead to suboptimal forecasting results. These findings underscore the importance of human oversight in the use of automated algorithms for time series forecasting and highlight the need for caution when employing automated methods. The ARIMA(1,1,1) model forecasts a flat trend (2023-2032) with production starting at 154,516 tonnes and declining slightly to 153,607 tonnes. This predicted stagnation implies that, with economic and population growth, soybean production will fall further behind domestic demand, leading to increased import reliance. These findings are crucial for farmers and policymakers as they can help inform decisions related to resource allocation, crop planning, and market strategies. Further analysis of these results is ongoing to gain deeper insights into the factors influencing soybean production trends in Turkey. Local producers could potentially benefit from increased production efficiency, improved competitiveness, and potential revenue growth by catering to both domestic and export markets. Additionally, understanding these trade dynamics can help stakeholders identify opportunities for collaboration or investment within the Turkish soybean industry.
Keywords: Soybean production, Turkey, Time series forecasting, ARIMA algorithm, NNAR, Auto-ARIMA
This study used the R statistical environment, version 4.2.2, developed by R Core Team (2022). The tidyverse meta-package, version 2.0.0, created Wickham et al. (2019), was employed for data manipulation and cleaning. For time series data extension, the tsibble package (version 1.1.3), developed by Wang, Cook, and Hyndman (2020), was utilized. To build forecasting models, the fable package (version 0.3.3) created by O’Hara-Wild, Hyndman, and Wang (2023a) was employed. For feature extraction and statistical analysis, the feasts package (version 0.3.1), developed by O’Hara-Wild, Hyndman, and Wang (2023b), was utilized. To create world maps, rnaturalearth version 0.3.4 by Massicotte and South (2023), rnaturalearthdata version 0.1.0 by South (2017),sf package version 1.0.14 and sp package version 2.1.2 with contributions from Pebesma and Bivand (2005) and Bivand, Pebesma, and Gomez-Rubio (2013), Pebesma (2018) were employed.
This analysis adapted and modified code from various sources, such as books, package manuals, vignettes, and GitHub repositories. The sources are cited as follows:
- Data preparing, manipulation, cleaning, and visualization: Wickham et al. (2019), Wang, Cook, and Hyndman (2020), Wang and contibutors (2024),
- Map Visualization: Massicotte and South (2023), South (2017), Pebesma and Bivand (2005), Bivand, Pebesma, and Gomez-Rubio (2013), Pebesma and contibutors (2024)
- Forecasting Models, Feature Extraction: Hyndman (2021), O’Hara-Wild, Hyndman, and Wang (2023a), O’Hara-Wild, Hyndman, and Wang (2023b), O’Hara-Wild and contibutors (2024)
Bivand, Roger S., Edzer J Pebesma, and Virgilio Gomez-Rubio. 2013. Applied Spatial Data Analysis with R, Second Edition. Springer, NY. https://asdar-book.org/.
Hyndman, R J. 2021. Forecasting: Principles and Practice. 3rd ed. Melbourne, Australia: OTexts.
Massicotte, Philippe, and Andy South. 2023. Rnaturalearth: World Map Data from Natural Earth. https://CRAN.R-project.org/package=rnaturalearth.
O’Hara-Wild, Mitchell, and contibutors. 2024. “Tidyverts/Fable.” 2024. https://github.com/tidyverts/fable.
O’Hara-Wild, Mitchell, Rob Hyndman, and Earo Wang. 2023a. Fable: Forecasting Models for Tidy Time Series. https://CRAN.R-project.org/package=fable.
———. 2023b. Feasts: Feature Extraction and Statistics for Time Series. https://CRAN.R-project.org/package=feasts.
Pebesma, Edzer J. 2018. “Simple Features for R: Standardized Support for Spatial Vector Data.” The R Journal 10 (1): 439–46. https://doi.org/10.32614/RJ-2018-009.
Pebesma, Edzer J, and Roger Bivand. 2005. “Classes and Methods for Spatial Data in R.” R News 5 (2): 9–13. https://CRAN.R-project.org/doc/Rnews/.
Pebesma, Edzer J, and contibutors. 2024. “Simple Features for R.” 2024. https://r-spatial.github.io/sf/.
R Core Team. 2022. R: A Language and Environment for Statistical Computing. Vienna, Austria: R Foundation for Statistical Computing. https://www.R-project.org/.
South, Andy. 2017. Rnaturalearthdata: World Vector Map Data from Natural Earth Used in ’Rnaturalearth’. https://CRAN.R-project.org/package=rnaturalearthdata.
Wang, Earo, and contibutors. 2024. “Tidyverts/Tsibble.” 2024. https://github.com/tidyverts/tsibble.
Wang, Earo, Dianne Cook, and Rob J Hyndman. 2020. “A New Tidy Data Structure to Support Exploration and Modeling of Temporal Data.” Journal of Computational and Graphical Statistics 29 (3): 466–78. https://doi.org/10.1080/10618600.2019.1695624.
Wickham, Hadley, Mara Averick, Jennifer Bryan, Winston Chang, Lucy D’Agostino McGowan, Romain François, Garrett Grolemund, et al. 2019. “Welcome to the tidyverse.” Journal of Open Source Software 4 (43): 1686. https://doi.org/10.21105/joss.01686.