Machine Leaning Methods
Multiple statistical methods can be used to set up a predictive model. In simple situations with linear relationships and only a few possible predictors, even linear models can be sufficient. Depending on the aim of the model, fast and simple methods like PLS can be used to get a rough idea of the prediction performance in process development. If the model should be used in real-time in a production process, more advanced non-linear methods such as Random Forest or Neural Networks should be considered. Here we give an overview of the advantages and disadvantages of some selected machine learning methods (Table 1).
Table 1: Most prominent statistical methods used for prediction of quality and quantity in bioprocess monitoring and control