Air-Access Stress as an Entrepreneurial Risk Signal: Interpretable Machine Learning for Tourism Enterprise Resilience in Philippine Island Gateways

Abstract

Tourism entrepreneurship in island destinations depends on reliable transport access because disruptions can quickly affect customer flows, bookings, cash flow, staffing, procurement, supplier commitments, and service delivery. This study evaluates whether abnormal air-passenger decline can function as an upstream entrepreneurial risk signal for tourism enterprises and whether interpretable machine learning adds decision value beyond simple operational rules. Civil Aviation Authority of the Philippines records for 2018-2025 were transformed into 714 observed gateway-months across eight tourism gateways, with blank cells retained as unobserved and explicit no-flight records coded as zeros. Models trained on 2018-2021 data were tested out of time on 2022-2025 observations against persistence and seasonal-naive baselines. Severe access-stress exposure was strongly negatively associated with tourism direct gross value added share of GDP (Spearman rho = -0.952, p < 0.001), while passenger growth was positively associated with TDGVA growth (rho = 0.929, p = 0.003); these annual associations are exploratory and pandemic-dominated. Random forest ranked next-month stress risk well (AUC = 0.940), but persistence achieved the higher F1 score (0.708 versus 0.400). SHAP showed that passenger scale, volatility, and momentum drove predictions. The findings support a hybrid entrepreneurial decision-support approach: machine learning for prioritization and onset surveillance, persistence for ongoing episodes, and local booking, revenue, occupancy, and operating indicators for action. The study positions interpretable AI as a managerial tool for tourism-enterprise resilience rather than as an end in itself.