ISSN(Online): 2320-9801 ISSN (Print) : 2320-9798 International Journal of Innovative Research in Computer and Communication Engineering (An ISO 3297: 2007 Certified Organization) Vol. 4, I ssue 5, May 2016 Copyright to IJIRCCE DOI: 10.15680/IJIRCCE.2016. 0405258 9531 Prediction of Landslide Displacement using NARX Model Talvinder Singh, Munish Kumar M.Tech Scholar, Dept. of IT, Centre for Development of Advanced Computing, Noida, Uttar Pradesh, India Senior Technical Officer, Centre for Development of Advanced Computing, Noida, Uttar Pradesh, India ABSTRACT: To prevent the damages due to landslide disaster, it is must to predict the changes in landslide displacement. Landslide displacement is direct reflection of state of the landslide. Landslide displacement time series and its influential factors could reflect the history of landslide displacement of dynamic system. In this paper we presented an application of NARX neural network for landslide displacement prediction. In addition to this, paper compares the performance of Radial Basis Function (RBF) network with NARX model. Displacement could be predicted by considering and reconstructing the relationship among the triggering factors and recurrent network such as NARX can reflect relationship among variables. Recurrent network with global feedback can learn the underlying dynamics of nonstationary environment. Comparison of RBF and NARX has been performed using MSE (Minimum Square Error). KEYWORDS: NARX neural network model; prediction of landslide displacement; back propagation learning through time I. INTRODUCTION Slope is a nonlinear dynamic system. There are various factors such as topography, groundwater, soil type, earthquake, rainwater are few parameters which trigger the landslide in a particular region. Landslides are the genetic type of slope and have same characteristics. Deformation of landslide can be predicted in short time by chaotic time series data recorded for any particular slope. Artificial neural network (ANN) model have an ability to recognize time series patterns and nonlinear characteristics, which gives better accuracy over the others methods, it become most popular methods in making prediction (Vaziri, 1997; Sharda, 1994; Jones, 2004; Toriman et al., 2009). Various network models have been adopted for predicting landslide, one such type of model based on RBF has been develop for prediction. In this research paper we have presented NARX neural network trained using back propagation learning algorithm using the time series data of landslide displacement which is discussed in section III for the early prediction of landslide displacement. The constructed network has been compared with the RBF network. II. NARX NETWORK In this paper, the architectural approach which we have proposed to deal with chaotic time series is based upon Nonlinear Autoregressive models with exogenous input, which are therefore called NARX recurrent neural networks. Recurrent network such as NARX incorporate a static multilayer perceptron or parts thereof [5]. They exhibit the nonlinear mapping capability of the multilayer perceptron. RN can have one or more hidden layers, because static MLP are often more effective than those single hidden layer. Each computation layer of a recurrent network has a feedback around it as illustrated in Fig 1. Let the vector x1(n) denote the output of hidden layer, x11(n) denote the output of the second hidden layer, and so on. Let the vector x0(n+1) denote the output of the output layer. Then the dynamic behavior of the recurrent network, in general, in response to an input vector u(n) is described by the following system of equations: x1(n+1) = φ1(x1(n), u(n)) x11(n+1) = φ11(x11(n), x1(n+1)) x0(n+1) = φ0(x0(n), xK(n+1) .. (1)