Analysis of the Impact of Public Holidays on Urban Tourism Passenger Flow Based on Causal Inference and Machine Learning

Under the context of China’s “holiday economy,” periodic festive events such as Labor Day and National Day trigger large-scale, short-term surges in tourism demand. Accurately assessing the true impact of public holidays is therefore of great significance for efficient resource allocation, optimized traffic management, and effective public service planning. However, existing forecasting methods typically treat public holidays as exogenous dummy variables or temporal indicators within time series or machine learning models. This approach struggles to disentangle the public holidays effect from confounding factors such as seasonal trends, weekend effects, and other concurrent external shocks, leading to relatively low prediction accuracy.To address this limitation, this study proposes a novel framework integrating causal inference with machine learning. First, the Synthetic Control Method (SCM) is employed to construct a counterfactual trajectory and quantify the Average Treatment Effect on the Treated (ATT) of public holidays. Second, the estimated causal effects are incorporated as dynamic features into an XGBoost model to build a causality-enhanced predictive model. Experimental results on a passenger flow dataset from Zhangye City, Gansu Province, demonstrate that the proposed causality-enhanced XGBoost model significantly outperforms the baseline models in predictive performance. By integrating causal inference with machine learning, this study substantially improves the accuracy of tourism demand forecasting.