[01] Carol Anne Hargreaves, Department of Statistics & Applied Probability, National University of Singapore, Singapore. [02] Loh Sheng Xiang, Department of Statistics & Applied Probability, National University of Singapore, Singapore.

In this paper, we utilize a Multi-Task Learning (MTL) approach to predict tourist expenditure, and compared its performance to other machine learning models. Using the MTL approach, different tasks representing different sub-categories of tourist expenditure was defined for our models, and based on the Root Mean Square Error (RMSE) and the Mean Absolute Error (MAE), the MTL approach had an upper hand in predicting unseen data compared to the Random Forest and Ridge Regression. We conclude that based on RMSE and the MAE, the Multi-Task Learning approach had a slight advantage over the Random Forest and Ridge Regression models, as the MTL approach was able to utilize the regularization term to facilitate learning and updating of weights from other tasks, thereby gaining an edge on its prediction power compared to the other Single Task Learning (STL) methods. Other than looking at the errors, we wanted to see whether the MTL approach was able to give a good interpretation of the model, such as which features were important in the prediction of expenditure. With regards to the interpretability of the MTL model, the MTL gave similar features of importance as the Ridge Regression model. For example, both models placed emphasis on the characteristics of the tourist’s accommodation for the prediction on total expenditure. Despite the Random Forest being a non-linear model, it seems that the MTL’s transfer learning ability outweighed the benefit of being a non-linear model. In conclusion, we have demonstrated that the MTL technique improved the prediction performance of the tourist expenditure.

Multi-Task Learning (MTL), Transfer Learning, Non-linear Model, Single Task Learning (STL), Prediction, Expenditure, Random Forests, Ridge Regression Model

[01] Emanuela Marrocu, Raffaele Paci, and Andrea Zara. “Micro-economic determinants of tourist expendi- ture: A quantile regression approach”. In: (2015). [02] Rob Law. “Demand for Hotel Spending by Visitors to Hong Kong: A Study of Various Forecasting Tech- niques”. In: (2000). [03] Ying Yu et al. “Statistical Modeling and Prediction for Tourism Economy Using Dendritic Neural Network”. In: (2017). [04] Boyu Wang and Joelle Pineau. “Online Boosting Al- gorithms for Anytime Transfer and Multitask Learn- ing”. In: (2015). [05] Rich Caruana. “Multitask Learning”. In: (1997). [06] Yu Zhang and Qiang Yang. “An overview of multi- task learning”. In: (2018). [07] Xavier Gilbert, Vishal M. Patel, and Rama Chellappa. “Deep Multitask Learning for Raiway Track Inspec- tion”. In: (2017). [08] Guangliang Gao et al. “Location-Centered House Price Prediction: A Multi-Task Learning Approach”. In: (2019). [09] Han Yuan et al. “Multitask learning improves predic- tion of cancer drug sensitivity”. In: (2016). [10] Juan L. Nicolau and Francisco J. Mas. “Heckit mod- elling of tourist expenditure: evidence from Spain”. In: (2005). [11] Frederick Dayour, Charles Atanga Adongo, and Fran- cis Taale. “Determinants of backpackers’ expendi- ture”. In: (2015). [12] Juan Gabriel Brida and Raffaele Scuderi. “Determi- nants of tourist expenditure: a review of microecono- metric models”. In: (2012). [13] Lingling Wu, Junyi Zhang, and Akimasa Fujiwara. “Tourism participation and expenditure behaviour: Analysis using a scotbit based discrete-continuous choice model”. In: (2013). [14] Woo Gon Kim et al. “Factors affecting the travel ex- penditure of visitors to Macau, China”. In: (2011). [15] Simon Rudkin and Abhijit Sharma. “Enhancing un- derstanding of tourist spending using unconditional quantile regression”. In: (2016). [16] Zheng Cao, Gang Li, and Haiyan Song. “Modelling the interdependence of tourism demand: The global vector autoregressive approach”. In: (2016). [17] Ping-Feng Pai, Kuo-Chen Hung, and Kuo-Ping Lin. “Tourism demand forecasting using novel hybrid sys- tem”. In: (2013). [18] Matthieu Kowalski. “Sparse Regression Using Mixed Norms”. In: (2009). [19] Andy Liaw and Matthew Wiener. “Classification and Regression by Random Forest”. In: (2002). [20] Gareth James et al. An Introduction to Statistical Learning. 7th ed. Springer, 2017. [21] Vijay Mahajun, Arun K. Jain, and Michel Bergier. “Parameter Estimation in Marketing Models in the Presence of Multicollinearity: An Application of Ridge Regression”. In: (1977). [22] Argyriou, A.; Evgeniou, T.; Pontil, M. Multi-task feature learning. In Proceedings of the 19th International Conference on Neural Information Processing Systems (NIPS’06), Vancouver, BC, Canada, 4–7 December 2006; MIT Press: Cambridge, MA, USA, 2006; pp. 41–48. [23] Michelucci, U, and Venturini, F. “Multi-Task Learning for Multi-Dimensional Regression: Application to Luminescence Sensing”, MDPI, Applied Sciences, (2019), 9, pp. 4748. [24] Zhang, Y.; Yang, Q. A Survey on Multi-Task Learning. arXiv 2018, arXiv: 1707.08114. [25] Thung, K. H.; Wee, C.-Y. A brief review on multi-task learning. Multimed. Tools Appl. 2018, 77, 29705–29725. [26] Liebel, L and Körner, M. “Auxiliary Tasks in Multi-task Learning”, Computer Science, Computer Vision and Pattern Recognition (2018). arXiv: 1805.06334 [cs.CV].