Time series prediction using focused time lagged radial basis function network

Abstract

In this paper temporal processing of time series function has been done using radial basis function network. Radial basis function network structure is actually static but it has been converted into dynamic one using memory component. Proposed dynamic radial basis function network is called as focused time lagged radial basis function network (FTLRBFN). In a time series function, output at any given instant of time depends on the past values of the inputs. This feature is exploited while implementing the FTLRBFN. Back propagation algorithm based on gradient descent principle is used to adjust the parameters of radial basis function network. The proposed FTLRBFN is also implemented to simulate the complex time series function. The results so obtained show that FTLRBFN is effective in approximating any complex time series function. Comparison in terms of average mean square error is also made when multi layer feed forward neural network (MLFFNN) is used in the proposed scheme. It is found that the proposed scheme with radial basis function network has given less average mean square error as compared to that obtained with MLFFNN in the scheme.