Scholarlink

<p>Assessing livestock resilience in tropical climates is often hampered by limited sample sizes, which constrains the use of powerful predictive analytics. This study introduces and validates a simulation-augmented machine learning framework to classify heat stress levels in Holstein × White Fulani crossbred dairy cows. The study evaluated forty-five lactating cows over 84 consecutive days, monitoring physiological indicators and urinary biomarkers, yielding 3,780 observations per parameter. To enable robust model training, the field dataset was synthetically expanded using Monte Carlo simulation to 1,000 observations. A Random Forest classifier trained on the augmented data successfully predicted heat stress categories (no stress, moderate, severe) with an accuracy of 89.30%. SHapley Additive exPlanations (SHAP) analysis identified respiration rate and urinary ammonia as the most influential predictors. While the cows exhibited adaptive responses like increased respiration (<i>r</i> = 0.39, <i>p</i> &lt; 0.001), their core body temperature remained relatively stable, validating the model’s focus on more sensitive indicators. These findings demonstrate that the framework can provide an accurate, early-warning system for heat stress, offering a powerful decision support tool for on-farm management and genetic selection of cows for thermotolerance.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Simulation-augmented machine learning characterization of thermotolerance in White Fulani crossbred dairy cows

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月19日

查看全文
  • Mahmood Aliyu,
  • Akeem Babatunde Sikiru,
  • Aliyu Haxy Dikko,
  • Stephen Sunday Acheneje Egena,
  • Olushola John Alabi,
  • Kasim Sakran Abass

摘要

Assessing livestock resilience in tropical climates is often hampered by limited sample sizes, which constrains the use of powerful predictive analytics. This study introduces and validates a simulation-augmented machine learning framework to classify heat stress levels in Holstein × White Fulani crossbred dairy cows. The study evaluated forty-five lactating cows over 84 consecutive days, monitoring physiological indicators and urinary biomarkers, yielding 3,780 observations per parameter. To enable robust model training, the field dataset was synthetically expanded using Monte Carlo simulation to 1,000 observations. A Random Forest classifier trained on the augmented data successfully predicted heat stress categories (no stress, moderate, severe) with an accuracy of 89.30%. SHapley Additive exPlanations (SHAP) analysis identified respiration rate and urinary ammonia as the most influential predictors. While the cows exhibited adaptive responses like increased respiration (r = 0.39, p < 0.001), their core body temperature remained relatively stable, validating the model’s focus on more sensitive indicators. These findings demonstrate that the framework can provide an accurate, early-warning system for heat stress, offering a powerful decision support tool for on-farm management and genetic selection of cows for thermotolerance.