Predictive analytics implementation in the logistic industry

Nowadays more than ever supply chain networks must meet increased demands. The digital transformation of the business is a necessity for the companies in the field of logistics in order to be able to increase their competitive advantages. During this transformation one of the most significant part plays the predictive analytics. Data driven decision making is crucial to supply chain activities. This however requires a more holistic view on implementation of predictive analytics in operational logistics processes. This paper aims to present possibilities for applying predictive methods in different operational processes in the context of a modern machine learning methodology framework and to demonstrate appropriate methods, techniques, algorithms, and software technologies. The scope of the research covers business processes in logistics organization. Several methodologies for design of predictive analytics frameworks have been evaluated and on this basis an adaptation of Microsoft Team Data Science Process is proposed. The methodology is demonstrated with an original practical implementation on a dataset provided by a logistics company. During each stage from the methodology suitable technologies, machine learning algorithms and evaluation measures have been applied. Conclusions are drawn regarding possibilities to implement the framework and to extract useful knowledge. Since the presented models are fitted to the used data set, the model explanation and interpretation is limited to the inherent data patterns and dependencies. Empirical results show that the best performing models are those trained with stacked ensembles and XGBoost algorithms. The model interpretation is implemented with SHAPley values and Partial Dependency Plots. The study is part of Project BG05M2OP001-1.002-0002-C02 "Digitalization of Economy in a Big Data Environment".