XAI-based Decision Support System for Heat Stress in Dairy Cattles

Heat stress is a major constraint to dairy production in humid subtropical regions, significantly reducing milk yield in primiparous Holstein cows in Mazandaran province, Iran. This study develops an XAI-based interpretable early warning framework to enable proactive management by integrating multi-method estimation of the upper critical Temperature-Humidity Index (THI) threshold with routinely collected farm records and real-time meteorological data. Daily milk yield data from ۲,۴۱۵ cows across three herds (۲۲,۷۱۲ records) were adjusted for days in milk, season, and herd effects to calculate percentage yield drops. Non-parametric methods, including generalized additive model derivatives, random forest split points, and XGBoost, yielded a consensus THI threshold of ۷۳.۳ (۹۵% CI: ۴۵.۶-۸۹.۳). Despite modest model fits (R۲ = ۰.۰۱۶-۰.۱۲۴) reflecting biological variability, the XAI-based system demonstrates practical utility in identifying elevated heat stress risk periods, facilitating timely interventions such as targeted cooling or feeding adjustments. The framework offers a scalable, transparent tool for enhancing resilience in dairy systems facing escalating climate challenges.

کلیدواژه ها:

xAI ، Heat Stress ، THI Threshold ، Dairy Cattle ، Climate Mitigation