Performance Analysis of Prediction Methods on Tokyo Airbnb Data: A Comparative Study of Hyperparameter-Tuned XGBoost, ARIMA, and LSTM

The rapid growth of the digital economy has increased the importance of accurately predicting Airbnb property occupancy rates, especially in dynamic and competitive markets such as Tokyo, Japan. Property owners face significant challenges in forecasting occupancy rates due to seasonal patterns, non-linear trends, and complex temporal dependencies within the data. Addressing these challenges, this study investigates the performance of ARIMA, XGBoost, and LSTM models in predicting Airbnb occupancy rates in Tokyo. The dataset is collected from Airbnb listings and includes relevant features such as location, price, customer reviews, and historical occupancy rates. The models were optimized using Grid Search for ARIMA and Random Search for XGBoost and LSTM to identify the best hyperparameter configurations. Evaluation metrics included Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Coefficient of Determination (R²), which are more appropriate for regression tasks. The results indicate that XGBoost achieves the highest R² (0.23), followed by LSTM (0.19) and ARIMA (0.03). However, the low R² values suggest that the models struggle to capture occupancy rate variations, indicating the potential influence of unmodeled external factors such as seasonality and policy changes. This study highlights the importance of hyperparameter tuning in improving prediction accuracy and contributes by providing an in-depth comparison of regression-based models for Airbnb occupancy forecasting.