Deep Sensitivity Analysis for Objective-Oriented Combinatorial Optimization

Abstract

Pathogen control is a critical aspect of modern poultry farming, providing important benefits for both public health and productivity. Effective poultry management measures to reduce pathogen levels in poultry flocks promote food safety by lowering risks of food-borne illnesses. They also support animal health and welfare by preventing infectious diseases that can rapidly spread and impact flock growth, egg production, and overall health. This study frames the search for optimal management practices that minimize the presence of multiple pathogens as a combinatorial optimization problem. Specifically, we model the various possible combinations of management settings as a solution space that can be efficiently explored to identify configurations that optimally reduce pathogen levels. This design incorporates a neural network feedback-based method that combines feature explanations with global sensitivity analysis to ensure combinatorial optimization in multiobjective settings. Our preliminary experiments have promising results when applied to two real-world agricultural datasets. While further validation is still needed, these early experimental findings demonstrate the potential of the model to derive targeted feature interactions that adaptively optimize pathogen control under varying real-world constraints.

Figures (5)

• Figure 1: Design diagram of Deep Sensitivity Analysis multi-label combinatorial optimization (DS). To predict food-borne pathogens (Salmonella, Listeria, and Campylobacter) in the first stage of the process, we build a multi-label multi-layer perceptron architecture ($\mathcal{M}$) based on farm practices. In our initial dataset, $m$ samples and $n$ features are represented as $X$ and $m$ samples and 3 pathogens are represented as $Y$. Samples are divided into $k$ train samples and $m - k$ test samples. As a next step, we will construct an explainable artificial intelligence (AI) based Deep Sensitivity Analysis (DS) neural network. We use the training sample ($\mathcal{T}$) as a reference dataset, and used the model $\mathcal{M}$ to predict the pathogen content. Similarly, the pathogen content for every sample in the original dataset is predicted using the $\mathcal{M}$ model. These two predictions are used to calculate the global sensitivity score, which is then used to build the Deep Sensitivity Analysis (DS) neural network, together with agricultural samples. In the final step of our study, we learn the global explanation of feature combinations through objective-based selection.
• Figure 2: Reduction of MDR based on pre-harvest agricultural practices. The results indicated that highly unbalanced, complex datasets required extensive analysis to interpret their feature associations to achieve optimization in the initial stages. However, when compared to dynamic programming (DP), DS significantly reduces Salmonella, Listeria, and Campylobacter MDR at earlier stages of poultry production.
• Figure 3: Reduction of pathogen presence based on post-harvest agricultural practices. Results indicate that compared to dynamic programming (DP) algorithms, our proposed approach provides promising results in combinatorial optimization. Moreover, it shows that explainable AI that provides feature combination explanations can improve the decision-making during combinatorial optimization problems.
• Figure 4: The top environmental variables learned from DS combinatorial optimization to reduce the risk of food-borne illness. This figure illustrates the pre-harvest practices and DS generated values that can help to reduce the presence of MDR on poultry farms. Indications from these recommendations can serve as a basis for further analysis and practical experimentation.
• Figure 5: The top poultry practices learned from DS combinatorial optimization to reduce the risk of food-borne illness. Figure presents post-harvest practices and DS generated feature values to help reduce pathogen contamination. Certain features produce a negative score, which indicates a reduction in pathogens in the absence of the factors.