Tourism Demand Forecasting and Implications on Sustainable Tourism in Sri Lanka – a Machine Learning Modeling Approach

Abstract

The study examined three key aspects of tourism demand in Sri Lanka. First, it explored the historical context, identifying factors that influenced tourism demand, such as previous years' tourist arrivals, macroeconomic conditions like inflation and exchange rates, and the effects of domestic and external shocks, including civil uprisings, ethnic conflicts, terrorist attacks, global health pandemics, and severe economic crises. The study quantified potential earnings losses from such events over the past five decades in Sri Lanka, using six different machine learning models and a time series model as a baseline. Secondly, using the same modelling approach, the study projected future tourist arrivals for the remainder of 2024 and the medium term from 2025 to 2027. Lastly, based on these projections, the study assessed future resource needs in the tourism industry, such as required room capacity and direct employment. It also addressed aspects of sustainable tourism by examining future demand at two popular tourist destinations in Sri Lanka: the Sigiriya rock fortress and the Yala National Wildlife Park. The machine learning models included in the analysis are Support Vector Regression (SVR), Support Vector Machines (SVM) Regression, K-Nearest Neighbor Regression (KNN), Extreme Gradient Boosting Regression (XGBR), Gradient Boosting Regression (GBR), and Multiple Linear Regression (MLR). Additionally, the study used the Seasonal Autoregressive Moving Average Model (SARIMAX), as the baseline model. The study found that a 12-month lag in tourist arrivals, macroeconomic indicators like inflation and exchange rates, and external and domestic events significantly influence monthly tourist arrivals in the country. Regarding external and domestic shocks, the study identified several key events that notably impacted potential tourism earnings. These include the July 1983 riots, the Eelam War I and subsequent JVP uprisings in the 1980s, Eelam War III in the 1990s, Eelam War IV in the 2000s, and the combined effects of the Easter Sunday attacks, the COVID-19 pandemic, and the ensuing financial crisis. These incidents were quantified in terms of the potential earnings lost from tourism. The study also projected short- and medium-term tourist arrivals using the same machine learning modelling approach. The study also examined the potential impact of forecasted tourist arrivals on four key tourism-related variables in Sri Lanka: room availability, the projected need for direct employees, the total daily visitor footprint at Sigiriya, and the daily vehicle requirement for safaris at Yala National Park.