[Paper Review] Interpretable machine learning applied to on-farm biosecurity and porcine reproductive and respiratory syndrome virus

Abagael L. Sykes, Gustavo S. Silva+5 more|arXiv (Cornell University)|Jun 11, 2021

Animal Virus Infections StudiesAgricultural and Biological Sciences96 references23 citations

TL;DR

This study develops MrIML-biosecurity, an interpretable machine learning toolkit that predicts porcine reproductive and respiratory syndrome virus (PRRSV) outbreak risk on swine farms using biosecurity practices and farm demographics. By applying local and global model interpretation, it identifies key risk factors—such as employee turnover, shared haul trailers, and proximity to other swine premises—enabling data-driven, farm-specific biosecurity benchmarking with actionable insights for disease prevention.

ABSTRACT

Effective biosecurity practices in swine production are key in preventing the introduction and dissemination of infectious pathogens. Ideally, biosecurity practices should be chosen by their impact on bio-containment and bio-exclusion, however quantitative supporting evidence is often unavailable. Therefore, the development of methodologies capable of quantifying and ranking biosecurity practices according to their efficacy in reducing risk have the potential to facilitate better informed choices. Using survey data on biosecurity practices, farm demographics, and previous outbreaks from 139 herds, a set of machine learning algorithms were trained to classify farms by porcine reproductive and respiratory syndrome virus status, depending on their biosecurity practices, to produce a predicted outbreak risk. A novel interpretable machine learning toolkit, MrIML-biosecurity, was developed to benchmark farms and production systems by predicted risk, and quantify the impact of biosecurity practices on disease risk at individual farms. Quantifying the variable impact on predicted risk 50% of 42 variables were associated with fomite spread while 31% were associated with local transmission. Results from machine learning interpretations identified similar results, finding substantial contribution to predicted outbreak risk from biosecurity practices relating to: the turnover and number of employees; the surrounding density of swine premises and pigs; the sharing of trailers; distance from the public road; and production type. In addition, the development of individualized biosecurity assessments provides the opportunity to guide biosecurity implementation on a case-by-case basis. Finally, the flexibility of the MrIML-biosecurity toolkit gives it potential to be applied to wider areas of biosecurity benchmarking, to address weaknesses in other livestock systems and industry relevant diseases.

Motivation & Objective

• To address the lack of quantitative, evidence-based benchmarking for on-farm biosecurity practices in swine production.
• To develop a methodology that quantifies and ranks the impact of individual biosecurity practices on PRRSV outbreak risk.
• To create an interpretable machine learning tool that supports farm-level and system-level risk assessment and biosecurity optimization.

Proposed method

• Trained multiple machine learning algorithms on survey data from 139 swine herds to classify PRRSV status based on biosecurity practices and farm demographics.
• Developed MrIML-biosecurity, a specialized toolkit extending the MrIML framework for interpretable machine learning in biosecurity.
• Applied global benchmarking to rank variables by overall importance in predicting PRRSV risk across farms and production systems.
• Used local model-agnostic interpretation (e.g., SHAP values) to quantify each farm’s unique risk contribution from individual biosecurity practices.
• Generated dispersal and waterfall plots to visualize the direction and magnitude of each variable’s contribution to predicted outbreak risk.
• Validated model performance using metrics including Matthew’s Correlation Coefficient (MCC), specificity, sensitivity, and AUC.

Experimental results

Research questions

• RQ1 Which on-farm biosecurity practices and demographic factors most significantly influence the predicted risk of PRRSV outbreaks?
• RQ2 How can interpretable machine learning be used to rank and quantify the impact of individual biosecurity practices on disease risk at the farm level?
• RQ3 To what extent do shared equipment, employee turnover, and proximity to other swine operations contribute to PRRSV risk prediction?
• RQ4 Can model interpretation reveal actionable differences between farms with and without recent outbreaks?
• RQ5 How does the MrIML-biosecurity toolkit support personalized biosecurity assessment and benchmarking across diverse production systems?

Key findings

• Fifty percent of the 42 analyzed variables were associated with fomite spread, while 31% were linked to local transmission pathways.
• Key risk factors included high employee turnover, shared haul trailers, proximity to public roads, and high local swine premises density.
• The number of visits from veterinarians and managers per month showed a negative contribution to risk, indicating a protective effect.
• Farms with PRRSV-positive animals hauled on cull breeding or wean animal trailers had significantly increased predicted outbreak risk.
• Local interpretation revealed that individual farms varied in which practices contributed most to their predicted risk, enabling case-specific biosecurity prioritization.
• The MrIML-biosecurity toolkit demonstrated strong predictive performance, with high MCC, specificity, sensitivity, and AUC, confirming its reliability for risk benchmarking.