Predictive Modeling for Tourist Arrivals: Assessing the Impact of Weather Data Using Machine Learning Models.

Abstract

This pioneering study explores the impact of incorporating weather data into the prediction of tourism demand in Sri Lanka, marking one of the first assessments of its kind in this context. We assessed the ability to forecast using three machine learning models—Long Short-Term Memory, Random Forest, and Support Vector Regression —using monthly data spanning January 2017 to October 2023. Two different sets of features were used to fit the models: one included only historical tourist arrival data, while the other included current and historical air quality data, which was the weather variable that had the greatest correlation with tourists arrival. Support Vector Regression demonstrated the best performance among all the models, with mean squared errors (MSE) of 0.01 and 0.16 for the models containing historical tourist arrivals and air quality as features, respectively. Interestingly, models using only historical tourist arrival data outperformed those incorporating weather data, suggesting that the inclusion of these external factors did not enhance forecasting accuracy as expected. These findings provide valuable insights for data-driven decision-making in the tourism sector, highlighting the need for careful feature selection and consideration of relevant external influences.