Predicting the next life event of cows by applying deep learning on sequential and pictorial data

In this research, we investigate whether a cow's future state such as calving, sickness or even death can be predicted based on its history of recorded fertility and disease events. In particular, we show how Markov for discrimination models (MFD) can estimate the posterior class probabilities and how deep learning algorithms such as Long Short Term-Memory networks (LSTM) can improve the classification accuracy. Additionally, we also investigate whether data augmentation with pictorial information can enhance the predictive performance even further by using convolutional neural networks (CNN). Results show that an ensemble model incorporating sequential as well pictorial information performs best and that the model is able to accurately predict future states such as calving, mastitis, pregnancy and death. The framework presented in this research can be used to enhance current animal monitoring systems with better animal welfare and higher financial returns for the farmer as a result.