CANTER: A Novel Causal Model for Tourism Demand Forecasting

Xin Han, Haiyang Xia, Ye Zhu, Gang Li, Rob Law

Abstract: Tourism demand forecasting is fundamental to strategic planning and economic policy, particularly given the sector’s substantial contribution to regional economies. While artificial intelligence (AI) models incorporating Search Engine Data (SED) have shown promise in tourism prediction, existing approaches often fail to capture the intricate causal dependencies among SED variables, thereby limiting their predictive capabilities. To address this limitation, we propose CANTER (Causal AI eNhanced Tourism dEmand foRecasting), a novel framework that leverages causal discovery and graph embedding techniques to integrate causal relationships among SED variables into AI-based forecasting models. The CANTER architecture comprises four key components: (1) temporal feature extraction from SED using sliding window analysis, (2) causal structure learning through the PCMCI algorithm, (3) causal graph representation learning via the Weave-Labelling model, and (4) integration of learned causal embeddings into predictive AI models. Empirical evaluation using Macau’s visitor arrival data demonstrates that CANTER achieves superior forecasting accuracy across multiple time horizons compared to baseline models. Our findings establish causal AI’s efficacy in forecasting tourism demand and provide methodological insights for advancing causality-aware data mining applications in tourism analytics.