Daily Rainfall Prediction Using One Dimensional Convolutional Neural Networks

Abstract

Rainfall is influenced by several factors, such as air temperature, humidity, and wind speed. But at this time, the weather conditions in the territory of Indonesia are increasingly uncertain, making them difficult to predict. The machine can learn the prediction of rainfall. This research proposed methods to predict rainfall within 14 days using ID Convolutional Neural Networks. Weather data were obtained from a weather observation station on Meteorological, Climatological, and Geophysical Agency (BMKG) website for ten years. First, data will be interpolated to fill in the missing value. Then the data will be segmented by overlapping and normalized to generalize the value of climate data to 0-1, including taking overlapping data to be an advantage in providing a linkage of sequential data values. Then the training and prediction process uses ID Convolutional Neural Networks. The accuracy generated using the Adam optimization model for testing data is 81.46%, and Loss is 0.0018.