ORCID

https://orcid.org/ 0000-0003-0177-2320

Department

Biological Sciences

Year of Study

2

Full-time or Part-time Study

Full-time

Level

Postgraduate

Presentation Type

Poster

Supervisor

Dr. Nóirín McHugh

Supervisor

Dr. Craig Murphy

Abstract

Animal live-weight contributes to profitability in beef herds and is a key determinant of overall efficiency of the beef sector. The objective was to develop a novel editing criteria for anomaly detection of beef cow and calf live-weight data. Live-weight data from five sources (i.e., professionally-recorded, owned-scales, borrowed-scales, scales hired from a depot, other) were available from the Irish Cattle Breeding Federation.

A number of alternative methods were used for anomaly detection including: generation of within-herd regression estimates, partial correlations between cow and calf live-weight records and mahalanobis distance. Across each method a value was calculated for each herd based on their distance from the assumed gold standard (i.e. professional recorded herds).

Results suggest anomaly detection should use both within-herd regression and partial correlation estimates jointly. A contingency table was formed with herds grouped into five classes according to their herd-level regression and partial correlation estimates between cow and calf live-weight. Based on mahalanobis distance 20% of herds were considered anomalies, however the data source did not impact data quality. Herd characteristics including herd size, calving season, month of live-weight recording, proportion of dams born within the herd of weighing, and proportion of dams of dairy breed origin per herd were shown to be associated with the quality of data recorded.

Results from this research will contribute to the formation of a specific editing criteria for live-weight data and assist herds with poor recording in improving data quality, thereby resulting in improved data submitted for use in national genetic evaluations.

Keywords:

data mining, beef, live-weight, anomaly detection

Start Date

June 2022

End Date

June 2022

Share

COinS
 
Jun 14th, 10:45 AM Jun 14th, 2:00 PM

Mining of producer recorded data; using beef calf and cow live-weight data as a case study

Animal live-weight contributes to profitability in beef herds and is a key determinant of overall efficiency of the beef sector. The objective was to develop a novel editing criteria for anomaly detection of beef cow and calf live-weight data. Live-weight data from five sources (i.e., professionally-recorded, owned-scales, borrowed-scales, scales hired from a depot, other) were available from the Irish Cattle Breeding Federation.

A number of alternative methods were used for anomaly detection including: generation of within-herd regression estimates, partial correlations between cow and calf live-weight records and mahalanobis distance. Across each method a value was calculated for each herd based on their distance from the assumed gold standard (i.e. professional recorded herds).

Results suggest anomaly detection should use both within-herd regression and partial correlation estimates jointly. A contingency table was formed with herds grouped into five classes according to their herd-level regression and partial correlation estimates between cow and calf live-weight. Based on mahalanobis distance 20% of herds were considered anomalies, however the data source did not impact data quality. Herd characteristics including herd size, calving season, month of live-weight recording, proportion of dams born within the herd of weighing, and proportion of dams of dairy breed origin per herd were shown to be associated with the quality of data recorded.

Results from this research will contribute to the formation of a specific editing criteria for live-weight data and assist herds with poor recording in improving data quality, thereby resulting in improved data submitted for use in national genetic evaluations.