Comparative Study and Feature Selection of Artificial Intelligence Models for Crop Yield Forecasting
Pekgöz, Eren (2025)
2025
Automaatiotekniikan DI-ohjelma - Master's Programme in Automation Engineering
Tekniikan ja luonnontieteiden tiedekunta - Faculty of Engineering and Natural Sciences
This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Hyväksymispäivämäärä
2025-09-01
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:tuni-202509018605
https://urn.fi/URN:NBN:fi:tuni-202509018605
Tiivistelmä
This thesis presents a comparative study of artificial intelligence (AI) models for crop yield forecasting, focusing on berry yields in the Nordic region. The research evaluates the performance of five AI models-Linear Regression (LR), Random Forests (RF), Multilayer Perceptron (MLP), Support Vector Regression (SVR), and Long Short-Term Memory (LSTM)-in predicting yields for blueberries, raspberries, strawberries, currants, gooseberries, and other berries of the genus Vaccinium. The study leverages climatic data, including temperature, precipitation, humidity, summer days, and frost days, to evaluate the accuracy and robustness of the models.
Key findings reveal that the most effective model varies by berry type: MLP performs best for blueberries, SVR for raspberries and strawberries, RF for currants and other berries, and LSTM for gooseberries. Although there is no single AI model that is the best for every type of berry, SVR or RF can be selected as the all-round reliable performer. Feature selection techniques, such as Pearson’s Correlation, LASSO Regression, Mutual Information, and Random Forest Importance, are applied to identify the most influential climatic variables, demonstrating that reduced feature sets can enhance model performance. The study highlights the importance of data preprocessing and the role of specific climatic factors in yield prediction.
The results provide practical insights for farmers and policymakers in the Nordic region, offering a framework to optimize yield forecasting models. The thesis concludes with recommendations for future research, including expanded data collection, hybrid modeling approaches, and the integration of explainable AI techniques to improve transparency and adaptability in agricultural applications.
Key findings reveal that the most effective model varies by berry type: MLP performs best for blueberries, SVR for raspberries and strawberries, RF for currants and other berries, and LSTM for gooseberries. Although there is no single AI model that is the best for every type of berry, SVR or RF can be selected as the all-round reliable performer. Feature selection techniques, such as Pearson’s Correlation, LASSO Regression, Mutual Information, and Random Forest Importance, are applied to identify the most influential climatic variables, demonstrating that reduced feature sets can enhance model performance. The study highlights the importance of data preprocessing and the role of specific climatic factors in yield prediction.
The results provide practical insights for farmers and policymakers in the Nordic region, offering a framework to optimize yield forecasting models. The thesis concludes with recommendations for future research, including expanded data collection, hybrid modeling approaches, and the integration of explainable AI techniques to improve transparency and adaptability in agricultural applications.