Comparative analysis of machine learning algorithms for estrous detection in dairy cows using sensor-based behavioral data across seasons.

Estrous detection is crucial for reproductive management in dairy herds, influencing artificial insemination efficiency, fertility rates, and herd productivity. This study aimed to find the best machine learning method to detect estrous in dairy cows using sensor data. Therefore, 14 Holstein and 5 Simmental cows at a farm in Ankara Province, Turkey were used. Using the CowManager SensO system, a total of 2.748 days of behavioral data were collected from cows, from which 124 estrous events were identified and recorded. In the study, the accuracy values of the CART, RF, MARS, SVM, and XGBOOST algorithms were found to be 0.68, 0.71, 0.75, 0.73, and 0.72, respectively, while the AUC values were determined to be 0.68, 0.71, 0.75, 0.73, and 0.72. The classification tree identified non-estrous cows based on activity level and estrous classification achieved when the active level was high-active ≥ 62 or when active ≥ 56 and ear temp < 26, resulting in 0.81 accuracy for estrous detection. As a result, the MARS algorithm outperformed the other algorithms in both accuracy and AUC values, highlighting its effectiveness for classification.