Publikationsdatenbank

Advancing Australian free-range egg production forecasts through machine learning:
a comparative model analysis (2024)

Art

Poster

Autoren

Adejola, Yusuf A.

Sibanda, Terence

Ruhnke, Isabelle (WE 15)

Iqbal, Amjad

Kearton, Tellisa

Pokhrel, Saluna

Welch, Mitchell

Kongress

113th PSA Annual Meeting

Louisville, Kentucky, 16. – 18.07.2024

Quelle

Poultry Science Association Annual Meeting Abstracts : — Poultry Science Association (Hrsg.)

— S. 194

Sprache

Englisch

Verweise

URL (Volltext): https://higherlogicdownload.s3.amazonaws.com/POULTRYSCIENCE/d925b046-3667-43aa-aeb6-5a36c497c07a/UploadedImages/2024_PSA_Pre-Meeting_Abstract_Book.pdf

Kontakt

Nutztierklinik: Abteilung Geflügel

Königsweg 63
14163 Berlin
+49 30 838 62676
gefluegelkrankheiten@vetmed.fu-berlin.de

Abstract / Zusammenfassung

Despite advancements in the free-range egg production industry, accurately forecasting laying rates and fluctuations remains a challenge. The study explores the efficacy of machine learning algorithms for forecasting production performance in free-range egg farms. In this study, we constructed six distinct predictive models across three experiments utilising historical production data, which includes variables such as laying rate, hen age, mortality rate, feed intake, water intake, and climatic weather data from four commercial free-range laying farms (Farm A, B, C and D) all of which utilise aviary systems. Through data cleaning and feature engineering, a robust dataset comprising 106 flocks and 35,346 individual days of data was obtained
using Python v3.12.2. The analytical framework was structured around three random forest experimental models for both regression and classification tasks: the first model was developed using single farm data (Farm A), the second model amalgamated data from multiple farms (Farm B, C, and D) and the third model integrated data from Farm A to D. Each model's predictive accuracy was subsequently evaluated against Farm A's production data, establishing it as the target farm or focal point of the case study. The random forest model was used to forecast laying rate and drops in laying rate (classified as drops vs normal production) using scikit-learn v. 1.4.1. Model performance was assessed using the Area Under the Receiver Operating Characteristic Curve (AUC) for classification tasks and the Root Mean Squared Error (RMSE) for regression tasks.
The first experiment, using data solely from farm A, demonstrated promising model performance, with a median AUC of approximately 0.86 for the classification task and an RMSE of around 2.8 for the regression task. This indicates the efficacy of farm specific historical data for accurate forecasting. In the second experiment, models exhibited considerable predictive power, with a median AUC of approximately 0.89 for classification and an RMSE of approximately 4.98 for regression, though slightly lower than the first experiment. The third experiment, employing a combined dataset from Farm A to D, yielded mixed results. While the regression task showed improved performance (RMSE of approximately 2.55), the classification task saw a
decline in performance (AUC of approximately 0.82). In conclusion, the research underscores the potential of machine learning models and analytics platforms to support production decisions in free-range egg farming. Future considerations include further exploring the practical applications of machine learning-driven decision support systems, such as the analytics dashboard.