Restaurants are frequent settings for foodborne illness outbreaks. Periodic inspection of restaurants is crucial to ensure commercial food establishments carry out safe food handling procedures. Predictive analytics can help identify problematic restaurants and maximize the utility of limited enforcement resources. To that end, I develop a machine learning model in this notebook. The resulting model can predict restaurants and other dining venues likely to face critical health code challenges in Las Vegas during the next inspection period.
The final model uses five numeric features to predict if a restaurant receives a C grade or below during the inspection (i.e., critical violation of sanitary practices). Three predictors refer to employee characteristics; the remaining two features describe the degree of violations. The model is a good tool if the purpose is to identify as many problematic restaurants as possible. The model could correctly identify 80% of problematic restaurants in a holdout test data set. At the same time, the model misclassified many compliant restaurants as non-compliant ones.