Monthly Tourism Demand Forecasting With COVID-19 Impact-Based Hybrid Convolution Neural Network and Gate Recurrent Unit

Abstract

The accuracy of tourism demand (TD) prediction, essential for managing available resources in the tourism industry, still needs to be improved with the unreliability of traditional algorithms. This research proposes a deep learning methodology that combines the convolution neural network (CNN) and gated recurrent unit (GRU), efficiently predicting Vietnam's tourism demand. The Pearson correlation coefficients are performed to nominate the most appropriate feature inputs. The proposed algorithm is analyzed and evaluated with other benchmark approaches, comprising the recurrent neural network (RNN), the long short-term memory (LSTM), the GRU, and the CNN. The experiments prove that the developed hybrid algorithm could outperform previous methodologies in predicting TD in some of Vietnam's provinces. The proposed algorithm could provide satisfactory predictions for tourism demand with a supreme enhancement of 77.1% MSE, 37.4% validating MSE, 46.0% MAE, 20.6% validating MAE, 76.6% MAPE, and 90.3% validating MAPE comparing across deep learning benchmarks.