Istanbul, as the world’s second-most visited city by international tourists, plays a critical role in global tourism. This study aims to identify the key factors contributing to the estimation of tourist arrivals in Istanbul across three periods between 2009 and 2024.
Tourist arrivals are forecasted using Gradient Boosting, XGBoost and Random Forest models. Predictors include average temperature, real effective exchange rate (REER), Google Trends, terrorist attacks and the Producer Price Index (PPI) for air transportation. Model performance is evaluated through root mean squared error and R², while SHAP analysis interprets the relative influence and direction of each variable.
XGBoost demonstrates the best overall predictive performance among the tested models. PPI, temperature and REER are the most influential predictors, while terrorist attacks have a limited impact. SHAP analysis highlights the positive role of Google Trends and temperature and shows that currency depreciation stimulates tourism by increasing affordability.
This study enriches tourism forecasting research by integrating economic, climatic and digital indicators within a machine learning framework. It demonstrates the value of nonlinear models and real-time data sources, such as Google Trends, for producing accurate, timely forecasts that can inform strategic tourism management and policy.
