Institutional Repository of Vilnius University
The importance of weather data in predicting agricultural gross value added using machine learning /
Abstract
dc:descriptionThe main objective of this thesis is to determine if weather variables, such as temperature and rainfall levels, can be employed for predicting the gross value added (GVA) in the agricultural sector. The literature review first motivates the research problem by summarising research on the impacts of weather on agriculture. Secondly, the review justifies the choice of machine learning (ML) methodology by comparing the advantages and limitations of conventional econometric methods with those of more recently adopted ML approaches. The objective of this thesis is achieved by training and interpreting machine learning models, namely XGBoost and LightGBM, to predict agricultural GVA on a granular NUTS-3 territorial division level, in the European Union. Firstly, the performance of models trained on weather data is compared against a benchmark model trained without weather data. Secondly, the models are interpreted using the Shapley additive explanations (SHAP) framework, to reveal 1) which weather variables are the most significant for predicting GVA growth, and 2) how specific values of weather variables impact the final predictions. The results of the analysis support the hypothesis that weather variables contain relevant information for predicting the agricultural sector’s GVA, as models trained on weather data outperformed the benchmark model. The model predictions could be improved with more advanced feature engineering and larger datasets, as their performance is moderate, and their interpretation reveals signs of potential overfitting. High temperatures coinciding with low rainfall are predicted to be detrimental to agricultural GVA, suggesting the need for investment in irrigation infrastructure. The relationship between some weather variables and predictions is found to be nonlinear, emphasising the necessity of methodology capable of capturing nonlinear relationships. These findings suggest that weather data could improve the accuracy of GVA forecasts, while applying causal machine learning methodology could enable informed and strategic climate adaptation policies in the agricultural sector.
Degree
thesis:*- Grantor dc:publisher
- Institutional Repository of Vilnius University
- Year dc:date
- 2024
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Byla, Titas,
Subjects
dc:subject × 1Rights
dc:rights- Statement dc:rights
-
- info:eu-repo/semantics/openAccess
- Language dc:language
- eng
Identifiers
dc:identifier.*- Repository record dc:identifier
- https://repository.vu.lt/VU:ELABAETD210642876&prefLang=en_US