Detecting heavy trucks from mobile phone trajectories using image-based behavioral representations and deep learning models

Abstract

This paper proposes an innovative methodology for detecting heavy trucks utilizing mobile phone data, addressing significant limitations inherent in traditional tracking methods, often characterized by high costs, intrusiveness, and incomplete data capture. By employing Call Detail Records (CDR) and introducing an image-inspired architecture, the study uses Convolutional Neural Networks (CNN) to model the microscopic behavioral patterns of mobile devices. Our numerical results show that our proposed approach outperforms more classical machine learning methods that rely only on aggregated features. This novel approach offers a scalable and cost-effective alternative to conventional methods, representing a pioneering application of image-based analytical techniques to mobile phone data within freight transport research. This work provides a robust tool for analyzing freight transport patterns, thereby supporting the development of strategies to mitigate the negative externalities of freight transportation while preserving its economic benefits.