Data Quality Management and Risk Assessment of Dairy Farming with Feed Behaviour Analysis Using Big Data Analytics with YOLOv5 Algorithm

Abstract

Dairy farming is crucial for global food security, providing essential products like milk and cheese. However, challenges such as animal health, economic instability, and environmental issues threaten the industry’s sustainability. This study utilizes big data analytics and machine learning, including YOLOv5 and Cascade Feedforward Neural Networks, to enhance feeding strategies, improve data quality management, and predict ketosis risks, ultimately improving cow health and preventing metabolic disorders. The study employs a combination of Apache Spark HDFS for handling large-scale data and YOLOv5 for real-time feed behaviour detection. Physiological data like rumination time, body temperature, and activity levels are collected, along with behavioural data from YOLOv5. These data types are integrated into a unified training pipeline, with the Cascade Feedforward Neural Network [CSFEM] for ketosis prediction. A Butterfly Optimization Algorithm [BOA]-guided stacking ensemble is applied to optimize model performance. The approach was implemented for efficient data processing and risk assessment. The proposed system achieved 99.8% accuracy, 99.2% precision, and 99.4% recall, effectively predicting ketosis and mastitis risks, showcasing the power of big data and machine learning in dairy farming. Future research could enhance model generalizability by incorporating diverse datasets, real-time monitoring, environmental sensors, and genetic data, and refining YOLOv5 for better real-world adaptability.