Forecasting container yard storage services in maritime logistics: An application to ports in Turkey

In this thesis, the future demand for container yard storage services at two Turkish seaports with high container-handling volumes (coded 9010 and 9020 and will be referred to by these codes throughout the thesis for confidentiality reasons) is forecast by combining time-series analysis with machine-learning techniques. A 44-month data set covering January 2021 to August 2024 is analysed; for every month six observations are recorded, capturing three service categories (storage – empty, import – full, export – full) and two container sizes (20 ft and 40 ft). The raw data were cleaned through missing-value checks, outlier control and categorical recoding, then converted to ARFF format compatible with the timeSeriesForecasting plug-in of WEKA 3.8.6 so that temporal dependence could be preserved in the modelling stage. During forecasting, Linear Regression, Decision Table and Random Forest algorithms were applied, and their performance was evaluated using the correlation coefficient (r) and Relative Absolute Error (RAE). Results show that the Random Forest algorithm produced the most accurate forecasts for both ports. The combination of high correlation and low error statistics indicates that the models successfully captured the complex patterns of trend and seasonality embedded in storage demand. These forecasts can help port managers optimise crane and yard scheduling, labour allocation and infrastructure investment on a data-driven basis. In addition, the thesis contributes new empirical evidence to the limited Türkiye-focused literature on container storage demand forecasting and demonstrates the practicality of machine-learning methods in logistics operations management.