Distribution-Level Case Study: Forecasting of Air Freight Delays

This chapter explains the air freight delivery process and how the data should be preprocessed. It introduces the machine learning algorithm parameters that can influence classifier performance. The chapter performs a quantitative and qualitative comparison between all classifiers, after classifier parameter selection. It reports the results of online classification. The chapter describes the air freight dataset used in this case study. The dataset was introduced by Metzger, who compared machine learning, constraint satisfaction and quality of service aggregation techniques to forecast air freight delays. The chapter mentions that the freight is consolidated at a specified airport. It describes the selected parametrization for each classifier. The chapter utilizes stratified k-fold cross-validation for honest classification performance assessment. It performs hypothesis testing to select the best classifier or the best set of classifiers to solve the air freight delay forecasting problem.