An intelligent modelling responsive to statistical changes yet refrain from noise is required to describe continuously evolving operating process. In this paper, the importance of applying a customized self-tuning algorithm to regulates the parameter setting in machine learning (ML) simulation, particularly genetic algorithm is demonstrated. The investigation is conducted with the multiple-input-multiple-output hydraulic system dataset feature selection benchmarking and several notable findings are obtained over the course of study. First, overfitting issue encountered by ML black box modelling can be reduce with feature selection optimisation. Next, a fine-tuned genetic algorithm as a function of fitness function increases prediction accuracy and reduce cross-validation losses compared to out-of-the-box deep learning. Finally, the trade-off between computation cost and ML interference power are non-trivial.
 

Feature selection,Genetic Algorithm (GA),condition monitoring,self-tuning