Algoritmos inteligentes aplicados à predição de atrasos em transporte aéreo

Abstract

The air freight cargo industry is increasingly facing operational challenges due to tougher global competition and higher service level requirements from customers. Machine Learning techniques are consequently being applied as an aviation supply chain risk management (SCRM) approach in order to predict delays, reduce operational uncertainty and reduce costs. The objective of the research is to evaluate whether the use of machine learning techniques contributes to a better prediction of air transport shipment delays in order to optimize the performance of the international supply chain control capacity. For that, different classes of algorithms are tested in the data mining phase of KDD (Knowledge Discovery in Databases): Support Vector Machine, Random Forest, Artificial Neural Networks and K-Nearest Neighbors. The Random Forest algorithm achieved the best result with accuracy of 86% in the cross validation test scenario after a combined class balancing procedure. Overall, the research also adds to the current literature as both transport and supplier data are used in a specific machine learning application.