Rainfall Predictive Approach for La Trinidad, Benguet using Machine Learning Classification

Abstract

Use of rain as a source of irrigation water presents an effective use of natural water resources. Predicting the occurrence of rainfall plays a major role especially in an agricultural area with untimely rainfall like La Trinidad, Benguet. For a more efficient irrigation scheduling, a reliable method for rainfall prediction is needed. This entails the adaptation and utilization of suitable prediction approaches and techniques. Various analytical approaches and methods are made available to develop new techniques to predict future possibilities. This study aimed to propose an approach in predicting the occurrence and non-occurrence of rainfall in La Trinidad, Benguet based on various historical weather parameters. Five machine learning classification algorithms were used to build the predictive models for the weather dataset namely: Fine Decision Tree, Linear Discriminant, Course K-Nearest Neighbors, Gaussian Support Vector Machines, and Neural Network. A poor choice of model cannot further improve the predictions. To choose between models, focus must be put on the appropriate evaluation metrics. Among the 5 models, results suggest that Course K-Nearest Neighbor gives the highest performance in all the evaluation metrics. Course KNN, with a good accuracy of 81.1% proves to be the best model to use in predicting rainfall in La Trinidad, Benguet. Course KNN model evaluation reveals that Machine Learning Classification can be adopted to predict the occurrence and non-occurrence of rainfall.