Regression Models

In the context of regression models, the following algorithms were employed:

1. Bayesian Ridge Regression
2. Kernel Ridge Regression
3. Stochastic Gradient Descent Regression
4. Regression Artificial Neural Network (ANN) with 6 dense layers and ReLU activation function.

Time Series Forecasting

The second approach involves time series forecasting, where the method of sliding window is employed to transform the unsupervised problem into a supervised one. Previous n timeseries observations are used to predict the vessel’s fuel consumption for the next point. The implementation is based on a Long Short-Term Memory (LSTM) model with 3 LSTM layers and 1 dense layer, using tanh as the activation function for the LSTM layers.