Rice Blast Disease Forecasting for Northern Philippines

Rice blast disease has become an enigmatic problem in several rice growing ecosystems of both tropical and temperate regions of the world. In this study, we develop models for predicting the occurrence and severity of rice blast disease, with the aim of helping to prevent or at least mitigate the spread of such disease. Data from 2 government agencies in selected provinces from northern Philippines were gathered, cleaned and synchronized for the purpose of building the predictive models. After the data synchronization, dimensionality reduction of the feature space was done, using Principal Component Analysis (PCA), to determine the most important weather features that contribute to the occurrence of the rice blast disease. Using these identified features, ANN and SVM binary classifiers (for prediction of the occurrence or non-occurrence of rice blast) and regression models (for estimation of the severity of an occurring rice blast) were built and tested. These classifiers and regression models produced sufficiently accurate results, with the SVM models showing a significantly better predictive power than the corresponding ANN models. These findings can be used in developing a system for forecasting rice blast, which may help reduce the occurrence of the disease. Key-Words: machine learning, artificial neural network, support vector machine, rice blast disease