Prediction of 24-h and 6-h Periods before Calving Using a Multimodal Tail-Attached Device Equipped with a Thermistor...

Area of Science:

• Animal Science
• Agricultural Technology
• Machine Learning in Animal Husbandry

Background:

• Accurate prediction of calving is crucial for optimizing livestock management and animal welfare.
• Existing methods for calving prediction often lack precision or are labor-intensive.
• Technological advancements offer new possibilities for real-time monitoring of pregnant cattle.

Purpose of the Study:

• To develop and validate predictive models for calving events using a multimodal tail-attached sensor.
• To assess the efficacy of these models under different housing systems (tethering and untethering).
• To identify key physiological and behavioral indicators preceding calving.

Main Methods:

• Utilized a multimodal tail-attached device to collect physiological (tail skin temperature) and behavioral (activity, lying time, posture changes, tail raising) data.
• Extracted 40 features from sensor data and incorporated days to expected calving date.
• Developed calving prediction models using a support vector machine (SVM) algorithm.
• Validated model performance using cross-validation on data from 63 pregnant cattle.

Main Results:

• Calving within 24 hours was predicted with high sensitivity (97% tethering, 93% untethering) and precision (80% tethering, 76% untethering).
• Calving within 6 hours was predicted with high sensitivity (91% tethering, 90% untethering) and precision (88% tethering, 90% untethering).
• Models demonstrated effectiveness across both tethering (tie-stall) and untethering (free-stall, individual pen) housing conditions.

Conclusions:

• Supervised machine learning models utilizing tail sensor data enable effective calving prediction.
• The developed models are robust and adaptable to various cattle housing environments.
• This technology presents a significant advancement for proactive livestock management and improved calving assistance.