Sustainable planning of developing tourism destinations after COVID-19 outbreak: a deep learning approach

ABSTRACT

Tourist destinations across the globe have been hit by the worst of the crisis that ensued the Covid-19 pandemic, and this concern is exponentially worse in developing countries. Sustainable planning of these countries to face the unheralded crises demands an approach to provide the most efficient solutions using past experiences, unique characteristics of the present crisis, and existing obstacles and challenges. The present study was conducted to develop a Deep Neural Network (DNN) model using experiences of different countries in beating the crises in the tourism industry. Relying on its generalization capability, the proposed model can forecast the ‘sustainable effectiveness of possible policies’ in developing countries by modelling the dynamics between the characteristics of these systems and the possible policies. A case study of a developing country was conducted to explain the model development process and its efficiency. Based on the data of the characteristics of the Covid-19 crisis and the tourism industry under study, the model outputs indicated that the most effective and sustainable policy to resume the pre-crisis conditions is to employ the combined policy of focusing on domestic tourism and crisis preparedness.

KEYWORDS:

• Deep learning
• developing tourism destination
• sustainable planning
• tourism planning and management
• covid-19 pandemic

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 The detailed description for how to input data in order to train and develop the model can be found in Section 4.2.

2 For more study, see Anthony and Bartlett (2009)

3 The Activation Function determines the behaviour of the node. Many types of activation functions are available in the neural network. Rectified Linear Unit (ReLU) and Tangent hyperbolic (Tanh) function as the activation function.

4 Regularization provides the magnitude of the weights, to the cost function. This method works as it simplifies the neural network’ architecture as much as possible, and hence reduces the possible onset of overfitting. Furthermore, the use of massive training data is also very helpful as the potential bias due to particular data is reduced.

5 Dropout trains only some of the randomly selected nodes in the neural network rather than the entire network. It is very effective, while its implementation is not very complex. In this context, some nodes are randomly selected at a certain percentage and their outputs are set to be zero to deactivate the nodes.