Multi-variable Model Of A Neural Network Based Weather Forecaster Using 2-stage Feature Selection
Abstract
This thesis proposes a novel approach for designing a neural network based forecaster that predicts more than one variable at a time. A second order two stage neural network training algorithm is used that employs orthogonal least square for training the output weights. In order to reduce the size of the network and train the forecaster optimally it uses time-domain feature selection and KLT transform based feature selection. The forecaster works well and the feature selection reduces the number of required inputs on the order of 70 %.