In this study a variety of machine learning techniques are utilized to develop models for the prediction of the main engine brake power and rate of revolution for bulk carriers. The focus is on reducing noise in input and output data by applying a spline smoothing method. As a result, a data preprocessing workflow incorporating spline smoothing is proposed to improve the generalization ability of the machine learning models. A thorough comparative analysis is performed on different methods, assessing their accuracy and generalization performance. The results, based on the smoothed data, show that the hypertuned Gaussian process regression model achieves the highest accuracy for both the validation and testing data. Additionally, linear regression models with smoothed data provide adequate accuracy for practical applications, leading to the creation of simple predictive formulas for brake power and rate of revolution, suitable for preliminary ship design.