Data analysis to predictive modeling of marine engine performance using machine learning

Modeling a complex marine engine system is always very challenging and often subjected to model uncertainties and time-varying inputs. A non-model based predictive modeling of marine engine system performance is therefore required. Predictive analytics tools such as Neural Network, Multiple Linear Regression, and Bagged Regression Tree Model are jointly used to model the marine engine system using different filtered input parameters. The unsupervised machine learnings such as Fuzzy C-Means amongst the K-Means Clustering and Self-Organizing Maps are used to reduce the root-mean-square error of the predicted model further by approximately 50%. With exploratory data analysis and applications of the proposed multiple learning schemes on the real-time engine data leads to a more robust and comparative approach to real-time engine model prediction for data engineers.