Predictability and forecasting automotive price based on a hybrid train algorithm of MLP neural network

Abstract

In this paper, using the Hurst exponent value H, we first show that the automotive price in Iran Khodro Company (IRAN) is predictable and therefore a good forecasting can be done using neural networks. We then introduce a new global and fast hybrid multilayer perceptron neural network (MLP-NN) in order to forecast the automotive price. In our new framework, we hybridize the genetic algorithm (GA) and least square (LS) method in order to train the connected weights of the network, which leads us to have a global and fast network. To do so, the connected weights between input and hidden layers are trained by GA and the connected weights between the hidden and output layers are trained by LS method. We finally apply our new MLP-NN to forecast the automotive price in Iran Khodro Company, which is the biggest automotive manufacturing in IRAN. The results are well promising compared with the cases when we apply the GA and LS individually. We also compare the results with the case when we employ the gradient-based optimization techniques such as Levenberg–Marquardt method as well as some heuristic algorithms such as extended tabu search algorithm instead of LS method and hybridization of MLP-LM with GA.