Self-adaptive parameter free artificial neural network training to model tourist arrival in Australian hotel industries

摘要

A self-adaptive and parameter free artificial neural network (ANN) training method to model seasonal accommodation demand in Australia using a small data set is presented in this study. The model uses tourist arrival pattern as time series data as well as other time series data such as CPI and GDP. The proposed ANN is trained using software developed by the author. This new ANN training method accelerates training epoch against the conventional back propagation training method. It avoids user-defined parameters; as a result, the training converges to the minimum trajectory of the ANN error space with less computational efforts. A constrained interpolation search repeatedly solves m number of reduced problems from E{sup}m to E{sup}1 space during training epoch. The trained ANN accommodation demand model, which is multivariate regression type provides close approximation to the actual data set. The training results enhance forecasting accuracy with mean absolute percentage error 0.2707 and mean percentage error -0.002, in training period and 1.03 and 0.516 in validation period respectively. The self-adaptive training method makes it easy to train an ANN. This feature is demonstrated in this study.