A dissertation submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy in Information and Communications Technologies, School of Engineering and Technology

Thesis (Ph.D.) - Asian Institute of Technology, 2020

In this dissertation rice yield prediction models based on historical datasets (1989-2017) for the main type of in-season rice cultivated in Thailand was studied. Models were generated using the following Machine Learning (ML) algorithms: Generalized Linear Model (GLM), Feed-Forward Neural Network (FFNN), Support Vector Machine (SVM), and Random Forest (RF). The prediction models were evaluated using Mean Absolute Percentage Error (MAPE), Root Mean Square Error (RMSE), and R 2 statistic. The results show that the FFNN outperforms the other models for rice yield prediction accuracy. In addition, the results are shown to be better than the existing prediction techniques including Linear Regression (LR), the k-Nearest Neighbor algorithm (kNN), Naive Bayes (NB), J48 a decision tree, RF, SVM, and Artificial Neural Network (ANN). The results confirm that the FFNN, a.k.a., Multi-Layered Perceptron (MLP), which is a deep neural network, can simultaneously account for complex nonlinear relationships in high-dimensional datasets. While the Big-O complexity and the execution runtime for training of the FFNN exceed the other models, its execution of predictions takes the least execution runtime. The practical implication of this study is to improve the quality of agricultural information dissemination services to farmers, wholesalers, retailers, middlemen, processors, financial institutions, policy makers, and the general public for the development of Thailand{u2019}s agricultural sector, rice supply chains and the economy as a whole.