Smart Forecasti̇ng of Hotel Occupancy via Neural Networks: Toward Data-dri̇ven Touri̇sm Management

Accurate demand forecasting is critical for strategic decision-making in the hospitality industry, particularly in culturally significant destinations where tourism patterns are often nonlinear and volatile. This study proposes a robust forecasting model based on a Nonlinear Autoregressive network with Exogenous Inputs (NARX), leveraging Artificial Neural Networks (ANN) to predict daily hotel occupancy rates. The model was developed using operational data from a 220-room hotel in Mardin, Turkey—a prominent heritage tourism hub. Instead of conventional macroeconomic or calendar-based variables, the model incorporates five strategically selected lagged occupancy values (y(t–1), y(t–3), y(t–7), y(t–15), y(t–30)) as exogenous inputs, enabling a lightweight yet effective forecasting architecture. With 336 daily observations from 2024, the dataset was split into training (70%), validation (15%), and testing (15%) sets. A single hidden layer with 25 neurons and 30 time delays was employed, and early stopping was used to prevent overfitting. The model achieved strong performance on test data, with a Mean Squared Error (MSE) of 51.17, Root Mean Squared Error (RMSE) of 7.15, and a correlation coefficient (R) of 0.8448. While residual analysis indicated some temporal autocorrelation, the model effectively captured daily trends. These findings highlight the potential of NARX-ANN models for accurate, real-time forecasting in dynamic tourism contexts, supporting improved pricing strategies, resource planning, and operational agility. The study also suggests further research on integrating external variables and advanced deep learning architectures to enhance long-term predictive power.