Accurate and reliable predictions of ship operating fuel expenditures can significantly increase the ship's operation environmental sustainability and profitability. Given there are general aims of shipping economically and reducing greenhouse gas (GHG) emissions worldwide, fuel consumption needs to be reduced to mitigate operational costs and GHG emissions. Improvement of operational strategies through accurately attributing ship fuel consumption rates to relevant ship operating modes is a way of achieving these aims. This, however, is difficult because the state of the vessel and its machinery systems are not constant (e.g., fouling extent and engine condition). Moreover, the state of the environment (currents, waves and winds) is also not constant. One commercial example where this challenge is particularly acute is in the case of distant fleet fishing operations, where fuel consumption often represents 50% or more of the total operational costs. In this industry there is a demand to develop a decision support system for optimal routing and planning. In this paper, these fishing operations are used to demonstrate a comparison of multiple regression algorithms for a fishing ship’s fuel oil consumption prediction model based on two in-situ vessel monitoring systems and environmental conditions forecast from public sources. Based on these data, the Correlation-based Feature Selection (CFS) method is carried out to select the best subset of predictive variables. Multiple regression algorithms are developed and applied, including Linear Regression, Random Forest, XGBoost and Neural Network with the result of Random Forest outperforming the rest of the algorithms for the two fishing vessels. The final selected models show accuracies of over 90% in all the speeds greater thanover 4 knots when the vessel is not in fishing-related operations, but searching for fishing grounds, which accounts for over 90% of the total fuel consumption. From the sensitivity tests carried out on the developed models, it was found that ship speed through water is the variable with critical importance for predicting fuel consumption in both of engine operating modes, which contributes to over 94.20% deviation to the baseline in kilograms per nautical mile, followed by month after last drydock (up to 4.34%) and environmental variables (up to 3.30%). This paper considers the practicalities of dealing with the complex data aggregation process from the two distinctly different sources, and demonstrates the relative performance merits of the different algorithms according to key indicators, such as the custom accuracy and the mean absolute error (MAE).