Machine learning strategies for the forecasting of pig growth in industry

Abstract

There is a growing demand to improve the sustainability of food production to improve food security whilst minimising any resulting impacts on the environment. This can, in- part, be achieved through the development of accurate pig growth prediction tools due to their potential in improving farm management practices. It is however necessary to ensure that the application of such tools is feasible on commercial farms by minimising their monetary costs. In this work, I explore several AI approaches for the forecasting of pig growth, each designed for different scenarios depending upon the nature of the data available. Firstly, the use of progressive technologies including Radio Frequency Identification (RFID) tagging systems and automatic feeding systems was explored. These systems were used to develop a novel recurrent neural network architecture for making accurate individual-level growth predictions in grower-finisher pigs. However, such technologies are too expensive to use on most commercial farms. Thus, in the second scenario I explored a more limited usage of technology where data was only collected at the birth and weaning of each pig. In this work, machine learning models were trained to estimate the finishing weight of each pig. In the final scenario, I sought to further reduce the reliance on expensive technologies by inferring individual-pig level growth trajectories without requiring RFID tagging to associate data with each pig. To achieve this, several methods were developed including the use of attention neural networks to estimate individual-level weight data from group-level weight data as measured by weighing platforms. The growth trajectories inferred through this were then successfully applied in growth forecasting. Throughout this thesis I describe numerous challenges within the pig farming industry and provide methods to tackle them using AI techniques. These methods provide practitioners with new approaches for managing pig farms whilst being easier to apply in a commercial environment compared to existing techniques.