An Analytics Prediction Model of Monthly Rainfall Time Series: Case of Thailand

Abstract

Rainfall prediction is regarded as a challenging task in an agricultural country like Thailand. A time series data especially rainfall and temperature needs analytics technologies to return a valuable knowledge. It has been recognized that a high accuracy of rainfall prediction model will be helpful for agriculturist and water management. The study area of this research is located in Thailand, which the daily rainfall and temperature time series data collected from five regions of Thailand were taken by Meteorological Department of Thailand from years 2000 to 2015. In this research, analytics method is proposed in the preprocessing steps, which are composed of data cleansing and data transform. Principal component analysis in feature selection step and weighted moving average are applied. In the prediction modeling, support vector regression (SVR) and artificial neural network (ANN) are employed. The results of the experiment showed the comparison of overall accuracy between ANN and SVR in five data sets over the area of study. The results of the experiment showed that the two prediction models gave a high overall accuracy, although SVR plays an important advantage in less computational time than ANN. This experiment is extremely useful not only as the most effective way to manage the amount of rainfall in water management for Thai agriculturist, but the proposed model can also become a representative in the monthly rainfall prediction model used in Thailand.