A Comparative Study of Different Curve Fitting Algorithms in Artificial Neural Network using Housing Dataset
Ali Al Bataineh, Devinder Kaur · 2018
Representing a given dataset with a mathematical model is a very useful tool in many engineering applications. Several techniques exist to evolve a mathematical model for a given data set. Non-Linear regression being the most often used. The curves can be generated from these mathematical models, which provide a visualization of how that model fits the data. In this paper, three algorithms to train artificial neural networks are used to develop a good fit for the data. The three algorithms are Levenberg-Marquardt, Bayesian Regularization, and Scaled Conjugate Gradient. These algorithms were applied to the housing data set. The comparative performance of these algorithms was compared using Mean Square Error (MSE), which represents the best curve fitting for these data sets. The mean squared error was computed for each algorithm. Levenberg-Marquardt had MSE of 7.0902. Scaled Conjugate Gradient at 15.2932, and Bayesian Regularization at 5.3480. It was found that Bayesian Regularization gave the best accuracy at 96.78%, followed by Levenberg-Marquardt at 94.53% and Scaled Conjugate Gradient at 90.51%.