eCommons

 

CHARACTERIZATION OF SENSOR AND NON-SENSOR COW, HERD MANAGEMENT, AND ENVIRONMENTAL DATA AND USE OF MACHINE LEARNING ALGORITHMS FOR PREDICTION OF PREGNANCY IN DAIRY CATTLE

Other Titles

Abstract

The overarching objective of the research presented in this thesis was to characterize associations between cow, herd, and environmental data with insemination outcome, and develop machine learning algorithms (MLA) to predict the outcome of the first service (FS) after calving in lactating dairy cows. The specific objective of the study presented in Chapter I was to compare patterns of multiple cow behavioral, physiological, and performance parameters collected by automated sensors before insemination for cows that became pregnant or not at FS. A secondary objective was to explore associations between pregnancy outcome at FS with previous gestation and early lactation performance and events, and with environmental conditions before insemination. An observational retrospective cohort study was conducted using data collected at a commercial dairy farm. Daily values for milk yield, milk components percent and yield, rumination and eating activity, physical and walking activity, resting time and bouts, body temperature, milk conductivity, and body weight collected by wearable and non-wearable sensors from -14 to 56 d after calving for 932 primiparous and 2,070 multiparous cows with a FS pregnancy outcome were available for analysis. Daily data were summarized as the average of seven periods of 4 to 7 d long from -14 to 56 d after calving and from -27 to -11, -10 to -3, -2 to - 1 d relative to timed AI for FS. The most notable differences observed for primiparous cows were greater milk yield, milk components yield, and fewer lying bouts per day for pregnant than non-pregnant cows. For the multiparous cow group, non-pregnant cows produced more milk and milk fat, had greater body temperature, more activity, more resting time, and had greater body weight changes after calving than pregnant cows. Associations of different strength and direction between FS outcome with previous gestation and previous and current lactation features, events, and performance for primiparous and multiparous cows were observed. Substantial variability between parity groups for the direction and magnitude of differences between pregnant and nonpregnant cows warrants use of parity either as a model predictor, or the development of parityspecific models for predicting FS outcome of lactating dairy cows. Chapter II of this thesis presents the development and performance of multiple MLA for predicting FS outcome using data presented in Chapter I. Decision Trees, Support Vector Machine, Logistic Regression, and Extreme Gradient Boosting models were built and evaluated for primiparous and multiparous only and for both parities combined. Overall, we observed that these MLA trained with a combination of automated sensor cow behavioral, physiological and performance data, as well as herd outcomes and environmental data presented a wide range of performance. The best performing algorithms (i.e., most performance metrics values in the 90 to 95% range) were those for primiparous cows using Support Vector Machine and Logistic Regression models. Overall, the performance of MLA for multiparous cows was poor (i.e., all performance metrics <70%) considering the implications of predictions for practical application. In conclusion, different supervised MLA trained with a combination of cow parameters collected by automated wearable and non-wearable sensors, herd outcomes, and farm environmental conditions, presented large variation in performance despite using the same input data and the same algorithms. Large variation in algorithm performance due to parity suggested that different models might have to be developed for predicting FS outcome for primiparous and multiparous cows. Further research is needed to identify a combination of predictors, methods to summarize input data from predictors, and develop procedures to train MLA that yield the level of performance required for practical use of algorithms in commercial farms.

Journal / Series

Volume & Issue

Description

157 pages

Sponsorship

Date Issued

2022-12

Publisher

Keywords

Dairy cow; Machine Learning; Parameters; Prediction; Pregnancy; Sensor

Location

Effective Date

Expiration Date

Sector

Employer

Union

Union Local

NAICS

Number of Workers

Committee Chair

Giordano, Julio

Committee Co-Chair

Committee Member

Nydam, Daryl

Degree Discipline

Animal Science

Degree Name

M.S., Animal Science

Degree Level

Master of Science

Related Version

Related DOI

Related To

Related Part

Based on Related Item

Has Other Format(s)

Part of Related Item

Related To

Related Publication(s)

Link(s) to Related Publication(s)

References

Link(s) to Reference(s)

Previously Published As

Government Document

ISBN

ISMN

ISSN

Other Identifiers

Rights

Rights URI

Types

dissertation or thesis

Accessibility Feature

Accessibility Hazard

Accessibility Summary

Link(s) to Catalog Record