Researchers at Institut Teknologi Bandung's School of Industrial Engineering have published compelling results showing that graph neural networks—a machine learning architecture designed for interconnected network data—can significantly outperform traditional operations research methods for supply chain optimization in archipelagic contexts. Their paper, accepted at the 2026 International Conference on Logistics and Supply Chain Management, demonstrates cost reductions of 18 to 23 percent in simulated last-mile delivery scenarios across eastern Indonesia.
The key insight driving the performance improvement is that traditional linear programming models treat each shipping leg as an independent optimization problem, missing the complex interdependencies created by Indonesia's hub-and-spoke maritime network. When a delay occurs at Makassar, it cascades through connections to Ambon, Ternate, and Jayapura in ways that static models cannot predict. The graph neural network approach, by contrast, learns the topology of the shipping network and can anticipate cascading effects, rerouting cargo preemptively before bottlenecks form.
The research team trained their models on five years of historical shipping data from Pelni and ASDP, supplemented by weather data from BMKG and port congestion records from the Ministry of Transportation. The training dataset encompassed over 2.3 million individual cargo movements across 147 ports, making it one of the most comprehensive maritime logistics datasets ever used for machine learning research.
Practical deployment faces hurdles. The model requires substantial computational resources for real-time optimization, and port operators in remote areas lack the IT infrastructure to integrate with such a system. The researchers are collaborating with a consortium of regional logistics companies to develop a lightweight deployment version that can run on edge computing hardware at major hub ports, with synchronization to a central model during off-peak hours. A field trial is planned for the Sulawesi-Maluku shipping lane in late 2026.
Dr. Sari Nirmala
Associate Professor, Industrial Engineering
Dr. Sari Nirmala researches AI and machine learning applications in logistics at Institut Teknologi Bandung, with a focus on emerging economy contexts.
