Pass or fail? Modeling restaurant inspection outcomes through interpretable machine learning
INTERNATIONAL JOURNAL OF HOSPITALITY MANAGEMENT, cilt.140, 2027 (SSCI, Scopus)
- Yayın Türü: Makale / Tam Makale
- Cilt numarası: 140
- Basım Tarihi: 2027
- Doi Numarası: 10.1016/j.ijhm.2026.104810
- Dergi Adı: INTERNATIONAL JOURNAL OF HOSPITALITY MANAGEMENT
- Derginin Tarandığı İndeksler: Social Sciences Citation Index (SSCI), Scopus, Hospitality & Tourism Complete, Hospitality & Tourism Index, Psycinfo
- Erciyes Üniversitesi Adresli: Evet
Özet
This study examines scores on U.S. restaurant health inspections through the lens of government data sources. Using over 160,000 inspections from the Chicago Department of Public Health's (DPH) Food Protection Program from 2010 to 2024, we explored categorical predictors of passing health inspections (e.g., type of inspection, risk rating, and number of re-inspections), as well as temporal predictors (e.g., month, day of the week, and year). In our estimates, we had to address the issue of class imbalance in the expected outcomes such as passing vs failing inspections, which we accomplished through logistic regression and CatBoost classification with weighting. We evaluated the accuracy of the models and used SHAP (SHapley Additive exPlanations) values for explainability, so we could identify which elements were most important in predicting passing health inspections. Our findings revealed that re-inspections were associated with a substantially higher likelihood of passing health inspections. We noticed comparable trends concerning the timing of inspections, whether across various seasons or on weekdays. Furthermore, we found that CatBoost and logistic regression offered different performance advantages. While logistic regression achieved slightly higher overall accuracy, CatBoost performed modestly better on several probability-based evaluation metrics and was particularly useful for capturing nonlinear patterns and supporting model interpretability. This study makes a methodological contribution to the field of regulatory compliance analysis by articulating an interpretable machine learning method. It also provides applicable contributions that demonstrate the usefulness of health code compliance to restaurant operators, public-facing policymakers, and the U.S. Health Departments that monitor food safety practices. The implications point to possible patterns in planner behavior regarding scheduling inspections and support the case for improving hospitality operations to strengthen compliance efforts.