Impedance spectroscopy for the characterization of swine reproductive states

Glass, Daniel J.

Permalink

https://hdl.handle.net/2142/106401 Copy

Description

Title

Impedance spectroscopy for the characterization of swine reproductive states

Author(s)

Glass, Daniel J.

Issue Date

2019-12-12

Director of Research (if dissertation) or Advisor (if thesis)

Bhalerao, Kaustubh

Committee Member(s)

• Grift, Tony
• Knox, Robert V

Department of Study

Engineering Administration

Discipline

Agricultural & Biological Engr

Degree Granting Institution

University of Illinois at Urbana-Champaign

Degree Name

M.S.

Degree Level

Thesis

Date of Ingest

2020-03-02T22:18:20Z

Keyword(s)

• Machine Learning
• Pig
• Pigs
• Swine
• Electrical Impedance Spectroscopy
• Impedance Spectroscopy
• Farrowing
• Estrus
• Animal Pregnancy

Abstract

This study uses machine learning to examine data acquired from a frequency sweeping electrical impedance spectrometer taking readings from sows in order to potentially determine if the resultant data can be used to predict a sow's date of estrus, date of farrowing, and if the inseminated sow has become pregnant. The frequencies analyzed were between 1000 Hz and 29000 Hz. The models were logistic regressions which were reduced by a stepwise Akaike Information Criterion. The data set for the fertility model included 250 frequency sweeps across 85 animals, for farrowing there were 921 observations across 421 animals, and for the pregnancy data set there were 395 observations across 227 animals. The fertility and farrowing detection models displayed substantial strength, with the latter holding particular promise in the face of conventional methods. The fertility model had a high Area Under Curve (AUC) of 0.946, while the two tested farrowing models had an AUC of 0.839 and 0.834. Pregnancy detection models were skewed by the retrieved dataset containing few non-pregnant animals, with only 50 observations across 29 non-pregnant animals as compared to 345 observations across 198 pregnant animals, and had an AUC of 0.839. Separating between late and early readings in the model is considered to be a potential improvement to the pregnancy model, with an AUC of 0.939, although additional observations, particularly of non-pregnant animals, are necessary. The groundwork was laid for further work to be done, particularly regarding whether an animal that has failed to become pregnant is going to return to estrus. Future models that incorporate more information about individual animals such as the number of litters previously birthed, as well as expanding this concept into other animals, are recommended.

Graduation Semester

2019-12

Type of Resource

text

Permalink

http://hdl.handle.net/2142/106401

Copyright and License Information

Copyright 2019 Daniel J. Glass

Owning Collections