Levenberg-Marquardt Deep Learning Algorithm For Sulfur Dioxide Prediction

Atmospheric pollutants play signification role in climate change as well as their great effect in human healthy. Prediction of such phenomena is very difficult due to the nonlinearity behavior of pollutant elements. Dynamic neural networks are good tools in dealing with such nonlinear problems that they can release the implicit dependencies in training data set through training algorithms for multilayer perceptron (MLP). The Levenberg-Marquardt algorithm (LM) is an epitome technique used to solve nonlinear problems. In this work MLP time series prediction with LM model will be build based on three years hourly data 2010:2012 cover great urban city Cairo, divided in three sets: 1500 target time steps for training (70%), over 450 target time steps for validation (20%) and over 225 target time steps for testing (10%). To avoid over fitting problem Levenberg-Marquardt algorithm stops training automatically when generalization stops improving, as indicated by an increase in the mean square error of the validation samples. Two performance measurements will be the methods of judgment the success of the proposed model: Mean Square Error (MSE) and relation coefficient (R). The proposed model is tested against recorded data set and proved superior.