تجزيه وتحليل عملكرد توليدمثلي گاوهاي شيرده در گله هاي بزرگ شيري با استفاده از الگوريتم هاي يادگيري ماشيني

Reproductive inefficiency remains one of the major challenges limiting productivity an‎d profitability in large-scale dairy herds. While numerous studies have applied machine learning techniques to animal growth an‎d production traits, most rely on controlled experimental datasets an‎d fail to address the heterogeneity an‎d noise inherent in real-world farm data. The present study aimed to identify key factors affecting three major reproductive indicators—days open (OD), days to first service (DFS), an‎d first-service conception success (FSPC)—an‎d to develop predictive models for these traits using machine learning approaches in Iranian Holstein dairy herds.Production, reproductive, an‎d management records were collected from 96 large industrial dairy herds across different regions of Iran over a 15-year period (2009–2024). After data cleaning an‎d quality control, 707,282 records from 282,524 cows were retained for analysis. A total of 25 predictor variables, including milk production traits, reproductive history, seasonal effects, management factors, an‎d calving characteristics, were used. Extreme outliers were removed using the ±3 stan‎dard deviation rule, an‎d the dataset was split into training (80%) an‎d testing (20%) subsets. XGBoost regression models were developed for DFS an‎d OD, while an XGBoost classification model was applied to predict FSPC. Model hyperparameters were optimized using GridSearchCV.Model performance was eva‎luated using RMSE, MAE, an‎d R² for regression tasks, an‎d accuracy, F1-score, AUC–ROC, an‎d confusion matrices for classification. The results demonstrated that factors such as number of inseminations, season of calving an‎d insemination, milk yield level, an‎d herd management practices played a decisive role in reproductive performance. Importantly, high-producing cows were shown to achieve acceptable reproductive outcomes, indicating that high milk yield an‎d reproductive efficiency are not necessarily antagonistic.This study demonstrates that machine learning models based solely on stan‎dard herd records already available on most dairy farms can accurately predict key reproductive indicators. The proposed framework provides a practical decision-support tool for dairy herd managers, enabling targeted reproductive management strategies to improve herd efficiency an‎d economic performance under real farm conditions. Keywords: Holstein dairy cow, reproductive performance, machine learning, first insemination success, open days, stan‎dard 305-day milk production, age at first calving, sperm origin, calving volume, livestock data analysis, data preprocessing

سعيد انصاري مهياري

مهرنوش فروتن

اميرحسين مهدوي , احمد ميره اي