Chalita Jainonthee. Application of data-driven tools for modeling and forecasting in food safety and food security situations. Doctoral Degree(Veterinary Science). Chiang Mai University. Library. : Chiang Mai University, 2025.
Application of data-driven tools for modeling and forecasting in food safety and food security situations
Abstract:
Food safety and security are critical concerns in veterinary public health, particularly regarding the production of animal-derived food products. This dissertation aimed to predict high rates of dead-on-arrival (DOA) incidents in meat-type ducks and to characterize beta-lactam antibiotic resistance patterns of Escherichia coli isolates from food-producing animals in Thailand. In the first study, a dataset comprising 18,643 transport records of meat-type ducks collected from 45 farms was analyzed. The predictive performances of several machine learning (ML) algorithms were compared, including Least Absolute Shrinkage and Selection Operator (LASSO), Support Vector Machine (SVM), Decision Tree (DT), Random Forest (RF), and eXtreme Gradient Boosting (XGBoost). Resampling techniques such as oversampling, undersampling, ROSE, and SMOTE were used to address data imbalance. The XGBoost model with oversampling achieved the highest predictive performance (AUC = 0.769, sensitivity = 0.642, precision = 0.476, and F1-score = 0.546). Key contributing factors identified included the number of ducks per truckload, lairage temperature at the slaughterhouse, and average body weight of ducks. These findings highlight the importance of physiological, logistical, and environmental factors in managing pre-slaughter mortality. In the second part of the study, antibiotic resistance patterns were examined in 3,219 E. coli isolates collected from poultry and swine at slaughterhouses and retail markets across different regions of Thailand. Hierarchical clustering was applied based on susceptibility to four beta-lactam antibiotics (ampicillin, cefotaxime, ceftazidime, and meropenem). The analysis revealed three distinct clusters. The first cluster consisted mainly of meat samples, reflecting similarities in processing environments. The second cluster comprised samples primarily from the Northeastern and Western regions, indicating regional differences in antibiotic usage. The third cluster included mostly cecal samples, suggesting resistance characteristics related to intestinal microbiota and antibiotic use at the farm level. Ampicillin resistance was high across all groups, exceeding 80 percent. Resistance to cephalosporins varied notably among sample groups, while meropenem resistance was rare, found in less than 2 percent of isolates. These results underline the need for region-specific surveillance and antimicrobial stewardship programs. In conclusion, this research demonstrates the application of advanced data analysis methods, including predictive modeling and hierarchical clustering, to support evidence-based decision-making. The findings provide practical insights for producers, veterinarians, and policymakers aiming to improve animal welfare, reduce pre-slaughter mortality, and manage antimicrobial resistance in Thailands livestock production systems in a sustainable manner.