Open Collections

Jonnatan, Livia

The steady growth of the world population also means the growing demand for food. Farms across the world aim to increase profitability and be more efficient, including dairy farms. In order to achieve that, farmers must be able to make the right decisions regarding when and how to use their resources. For dairy farmers, they are interested in knowing estrus events, which is an ideal time to inseminate and treat cows. Farmers traditionally use their expertise and experience to make these decisions, but this takes time and is hard to generalize and scale to larger herds. Nowadays, there are deployments of animal activity sensors, which help farmers gain more insights of their farms. This opens an opportunity to utilize machine learning algorithms to timely and accurately detect estrus events. The aim of this study is to analyze step count data from 332 cows to be used as a predictor of estrus events using several supervised machine learning algorithms, such as classification tree, K-Nearest Neighbors (KNN), Logistic Regression, and Linear Discriminant Analysis (LDA). The results reveal high accuracy, but low precision, recall, and F1-score. Although during estrus cycles there were increased physical activity for some cows, this is not generalizable for all cows. Using only the step count data has been shown to be insufficient to accurately predict estrus cycles.

Text

University of British Columbia. DATA 449

Vancouver : University of British Columbia Library

10.14288/1.0432068

Science, Irving K. Barber Faculty of (Okanagan); Computer Science, Mathematics, Physics and Statistics, Department of (Okanagan)

Undergraduate

DSpace