Advancing digitalization and high computing power are drivers for the progressive use of machine learning (ML) methods on manufacturing data. Using ML for predictive quality control of product characteristics contributes to preventing defects and streamlining future manufacturing processes. Challenging decisions must be made before implementing ML applications. Production environments are dynamic systems whose boundary conditions change continuously. Accordingly, it requires extensive feature engineering of the volatile database to guarantee high generalizability of the prediction model. Thus, all following sections of the ML pipeline can be optimized based on a cleaned database. Various ML methods such gradient boosting methods have achieved promising results in industrial hydraulic use cases so far. For every prediction model task, there is the challenge of making the right choice of which method is most appropriate and which hyperparameters achieve the best predictions. The goal of this work is to develop a method for selecting the best feature engineering methods and hyperparameter combination of a predictive model for a dataset with temporal variability that treats both as equivalent parameters and optimizes them simultaneously. The optimization is done via a workflow including a random search. By applying this method, a structured procedure for achieving significant leaps in performance metrics in the prediction of hydraulic test steps of directional valves is achieved.