Abstract

Two models of neural network (NN), namely a feed forward neural network (FFNN) and time delay neural network (TDNN) were proposed to model and predict daily rainfall in Central Java, Indonesia. TDNN used in this paper is feed forward networks with finite impulse response (FIR) filter on the input layer and hidden layer. There are some indicators which are frequently used to measure the accuracy of goodness of fit model and forecasting. Those indicators are mean square error (MSE) and Bayesian Information Criterion (BIC). Those two models would be applied to daily rainfall data at three locations: Napen, Pabelan and Klaten station. Data are split into two parts, namely training and testing data. The result shows that TDNN has better performance, due to its fewer parameters and prediction accuracy, than FFNN.

Original languageEnglish
Pages (from-to)193-201
Number of pages9
JournalInternational Journal of Applied Mathematics and Statistics
Volume44
Issue number14
Publication statusPublished - 2013

Keywords

  • Daily rainfall
  • Feed forward neural network
  • Multivariate time series
  • Time delay neural network

Fingerprint

Dive into the research topics of 'On the multivariate time series rainfall modeling using time delay neural network'. Together they form a unique fingerprint.

Cite this