Multi-algorithmic approach for detecting outliers in cattle intake data

• DOI: 10.1016/j.jafr.2024.101021
• Publication Date: 2024-01-28T05:31:55Z
• Authors: Jae-Min Jung; Dong-Hyeon Kim; Hyunjin Cho; Mingyung Lee; Jinhui Jeong; Dae-Hyun Lee; Seongwon Seo; Wang-Hee Lee
• Keywords: Weighted voting; Statistic

Monitoring cattle feed intake is crucial for evaluating animal health, productivity, and farm profitability. In particular, an abnormal related to the activity. Therefore, outlier detection forms basis monitoring. This study employed multiple algorithms ranging from statistics deep learning detect outliers in time-series data of intake. We used five models implementing mean + standard deviation, moving average, box plot, time series decomposition, autoencoder, attempted enhance performance by a voting system combine more than one model. Both plot decomposition demonstrated high accuracy (over 95 %) F1-score (harmonic precision recall). Thus, it reliably distinguished normal values outliers. Moving average exhibited true-skill statistic (TSS), thereby rendering suitable detection. The gave F1 TSS scores 0.49 0.65, respectively. enhanced compared with individual These results demonstrate that metrics vary depending on type algorithm. This, turn, highlights need select adapted monitoring objectives. algorithmic selection can be complemented system. demonstrates its potential generating reliable database accurate aiding decision-making livestock producers.