Natural Disaster Predictions of Whirlwinds in Cilacap Region of Central Java Using Adaptive Neighborhood Modified Backpropagation (ANMBP)

Abstract

BNPB’s data shows that in the last 10 years, the province of Central Java was the most whirlwind region if compared with other provinces in Indonesia which amounted to 1281 incidents and whirlwinds most often in Cilacap which amounted to 202 incidents. The purpose of this research was to predict whirlwind in Cilacap region of Central Java using meteorological parameters as predictive parameters and adaptive neighborhood modified backpropagation (ANMBP) parameters as a prediction method. In the preprocessing process, the input data variable was reduced from seven input variables to five input variables by using the principal component analysis (PCA) method. The prediction process using ANMBP, and trial and error processes are carried out repeatedly by changing the hidden layer values, learning rate, and the amount of training and testing data. The best performance is obtained from the distribution of training and testing data at 60–40%, it produced an MSE value of 0.0004, and accuracy is 85.59% using two hidden layers with values of 110 and 95 and learning rate of 0.1.