Decisões inteligentes em cidades complexas: integração de Machine Learning em sistemas de suporte à decisão

Abstract

In contemporary cities, urban logistics represents a critical challenge due to the growing demand for rapid deliveries, traffic congestion, and increasing pressure for sustainable practices. This environment reveals fragmented data, lack of standardization across available databases, and limited digital maturity within public institutions. At the same time, Machine Learning (ML) techniques provide new possibilities for transforming dispersed information into actionable knowledge, although a considerable gap remains between technical potential and practical implementation in urban management. Within this context, the research aimed to develop a theoretical framework that integrates ML into Decision Support Systems (DSS), with the purpose of supporting public managers in decision-making processes related to urban logistics. The study adopted the Design Science Research (DSR) methodology, combining systematic literature review, in-depth interviews, a structured questionnaire, and validation interviews. Ten specialists from the fields of Artificial Intelligence, innovation, and public management participated in the research, enabling triangulated analysis between qualitative and quantitative data. The findings demonstrated the feasibility of an eight-stage cyclical model designed to transform urban data into well-founded decisions. The study also identified institutional barriers, digital maturity requirements, and strategies for bridging data, algorithms, and management practices, reinforcing the practical applicability of the model to the public sector. The framework contributes to the literature by articulating technical, institutional, and operational dimensions of urban logistics, offering a consistent and adaptable conceptual basis. From a practical perspective, it provides clear guidelines for public managers facing fragmented data and the urgency for more agile and evidence-based decisions. As limitations, the research highlights the absence of real-world testing and the restricted sample size, opening opportunities for further studies in different institutional and territorial contexts.