57 Leveraging bioelectrical impedance analysis and machine learning for early detection of parasitic infections in dairy goats
Abstract Parasitic infections in livestock, particularly gastrointestinal nematode infections, pose a significant threat to animal health and productivity, leading to economic losses in dairy goat operations. Early and accurate detection of parasitic infections is crucial for mitigating these impacts and preventing anemia and weight loss in infected animals. Current diagnostic methods, such as fecal egg counts (FEC) and FAMACHA scoring, are time-consuming, labor-intensive, and require skilled personnel, limiting their practical application in large-scale operations. To address this challenge, bioelectrical impedance analysis (BIA) presents a promising alternative as a rapid, non-invasive tool for assessing animal health by measuring the electrical properties of tissues. Combining BIA measurements with machine learning algorithms can enhance the accuracy of detecting parasitic infections and predicting anemia severity in dairy goats. This study collected BIA parameters, including electrical resistance and reactance, packed cell volume (PCV), FEC, and FAMACHA scores from dairy goats housed at the USDA rRsearch Station in Byron, GA. A comprehensive machine learning pipeline was developed to evaluate the predictive performance of Support Vector Machines (SVM), Backpropagation Neural Networks (BPNN), and Random Forest models for regression and classification tasks to predict relative FAMACHA score based on previously mentioned parameters. Regression models yielded varying performance levels, with Random Forest achieving the highest R² score of 0.7287, indicating a strong predictive capability compared to BPNN (R² = 0.2512) and Support Vector Regression (SVR, R² = 0.2326). Classification models demonstrated improved accuracy, with Random Forest achieving a notable accuracy of 78.74%, outperforming BPNN (54.90%) and SVM (52.65%). These results suggest that BIA-derived parameters (resistance and reactance), particularly when analyzed using Random Forest algorithms, hold significant potential for predicting and classifying parasitic load and anemia-related conditions in dairy goats. The correlation analysis revealed a non-significant negative association between FAMACHA scores and BIA electrical resistance (r = -0.08) and reactance (r = -0.11), but a strong negative association between PCV and FAMACHA scores (r = -0.93), suggesting the critical role of PCV as an indicator of anemia severity. The FEC values showed moderate positive correlations with FAMACHA scores (r = 0.41) and goat parasitic infection level (r = 0.45), indicating that parasitic loads significantly influence health outcomes. Using machine learning models in livestock health monitoring presents a promising avenue for improving early detection and management strategies for parasitic infections. Future work will integrate a larger sample size and additional physiological and environmental factors to enhance predictive accuracy and provide a holistic decision-support tool for dairy goat management.
Read more