Predictive Model for Assets Control in a Peruvian Poultry Farm Applying Machine Learning

The poultry industry faces problems in bird health care since many environmental factors directly affect bird welfare and, therefore, production since birds are the driving force of the business. To maintain this welfare, birds are monitored and information from production is analyzed. Likewise, this monitoring in many cases only evaluates production without considering bird mortality. For this scenario, a solution called Farmsense that incorporates ML is presented and is where this work is focused. The study proposes to implement 2 prediction data-driven models based on ML for bird “mortality” and egg “production”. In order to train the models, a process was carried out where business data was taken based on previous studies, the relevant data was evaluated to train the proposed models, and the training of the regression models was carried out with the Random Forest and XGBoost algorithms. The validation of the results was done based on the <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{R}^{\mathbf{2}}$</tex>, MAE, MSE and MAPE metrics to find the best-fitting models. The results for the “production” prediction models showed good results obtaining MAPE values less than 10 % and <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{R}^{\mathbf{2}}$</tex> values greater than <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{8 0 \%}$</tex>, being the “production” model with Random Forest the best evaluated. In the future, with more data, it is possible to improve these values and keep the accuracy of the predictions when they reach the desirable values.