Comparative study of machine learning techniques to predict fuel consumption of a marine diesel engine

The motivation of this study is to compare four different machine learning algorithms which are support vector regression, multiple linear regression, J48 pruned tree, and M5 Rules, to predict the fuel consumption (FC) of a large marine diesel engine utilized as the main engine on a tanker vessel. This study aims to fill a literature gap by comparing two algorithms, which have not been used for this problem in the rule and tree-based literature, with the other two frequently used algorithms. The data gathered from noon reports and the logbook of an oceangoing tanker vessel involves the operational and environmental parameters. The model performances, prediction accuracy, and error deviations on the test set are demonstrated. The importance of each feature on fuel consumption is discussed regarding the rules created by the algorithms. The M5 Rules algorithm has the highest performance with the truest predictions with a correlation score of 0.9666, mean absolute error of 2.3536, and root mean squared error of 3.3947. Slip, speed, distance, and wind direction are the operational and environmental dependent variables that have more influence on the FC. M5 Rules algorithm has provided a clear sorting for each feature's importance regarding conditions.