Master's Thesis

Estimation of arrival times for shipping

A machine learning model trains the estimation of arrival times (ETA) in shipping on data from the Danish Automatic Identification (AIS) System

Pipeline:

1. Data preprocessing
   1. Identification of drives per port and MMSI
   2. Handling NaN, undefined, One-Hot-Encoding
   3. Automatic label generation
2. Model training from scratch for ports
   1. Architecture InceptionTime + Dense
   2. One model for each port (base-model)
3. Transfer of models to each other port
   1. Re-Train certain layers
   2. Iterate with different layers and less data
4. Evaluation (MAE, plots)

Preview:

MAE of normally trained and transferred models

• Base-Traning on diagonal
• Transfer from Port X to Port Y