Autor(es): Martins, João P. ; Ramos, Filipe R. ; Pereira, M. Teresa ; Oliveira, Marisa ; Ferreira, Fernanda A.
Data: 2025
Identificador Persistente: http://hdl.handle.net/10400.22/30189
Origem: Repositório Científico do Instituto Politécnico do Porto
Assunto(s): Logistics; Process modelling; Forecasting; Exponential smoothing; Deep learning; BJ-DNN model; Prediction error
The transportation of goods is critical to supply chains, directly influencing efficiency, cost management, and competitiveness. Accurate freight cost forecasting is essential for decision-making, enabling businesses to allocate resources effectively, reduce financial uncertainties, and ensure timely deliveries. This study, conducted in the After-Sales department of a global company, aimed to analyse freight costs per shipment and develop a predictive system based on predefined parameters. Historical data were examined using analytical techniques and time series metrics to identify suitable forecasting methodologies. Specific algorithms, including classical methodologies (exponential smoothing models) and hybrid deep learning models (BJ-DNN model), were tested to evaluate predictive accuracy. Results showed prediction errors ranging from 17% to 56% for exponential smoothing models and from 5% to 27% for BJ-DNN models, demonstrating the superior performance of hybrid approaches. These findings emphasize the potential of predictive models to enhance freight cost forecasting, minimizing error margins and optimizing resource allocation. This research provides a foundation for refining these methodologies, contributing to improved freight cost management and operational efficiency.
