A STEP TOWARDS IMO GREENHOUSE GAS REDUCTION GOAL: EFFECTIVENESS OF MACHINE LEARNING BASED CO2 EMISSION PREDICTION MODEL

Ships are the world's most economical means of freight transportation, and day by day, it is expanding quickly. The increase in ship transportation activities has resulted in a significant concern about CO2 emissions. International Maritime Organization has agreed to set a goal of reducing the maritime sector's total gas emissions by at least 50% by 2050. In this regard, a CO2 emission prediction model followed by an emission inventory can play a vital role in decision -making to optimize the ship's speed, draft, trim, and other influencing parameters under Ship Energy Efficiency Management Plan to decrease carbon emissions during operation. Machine learning, a branch of the data science approach, can be utilized to create effective emission-prediction models. In this research, two machine-learning models have been developed using actual voyage data collected from the noon reports of ships in Bangladesh. The models have been trained with the ship's speed, engine rpm, wind force, and sea condition during voyages. The models' performances have been assessed employing the Coefficient of Determination (R2) and Root Mean Square Error (RMSE). The prediction accuracies for the K Nearest Neighbor Regression model and the Light Gradient Boosted Machine Regression model are 84% and 81%, with RMSE of 5.12 and 5.53, respectively.