A team of researchers led by Professor Shinya Inazumi from Shibaura Institute of Technology in Japan has developed a groundbreaking machine learning approach to create highly detailed earthquake risk assessment maps for Tokyo. The new method produces 3D liquefaction hazard maps at a 200-meter grid scale—far more precise than traditional maps, which typically operate at 500 meters or coarser. This advancement is critical for a city like Tokyo, located in a seismically active region where soil liquefaction poses a major threat during earthquakes. Soil liquefaction occurs when strong shaking causes saturated, loose soils to lose strength and behave like a liquid, leading to building collapses and infrastructure damage. Major earthquakes such as the 1995 Great Hanshin-Awaji, the 2011 Great East Japan, and the 2024 Noto Peninsula events have all demonstrated the destructive power of this phenomenon, particularly in reclaimed coastal areas and river floodplains. Existing hazard maps often rely on limited borehole data and basic geostatistical models, resulting in coarse resolutions that fail to capture the fine-scale variations in soil conditions across Tokyo’s complex urban landscape. To address this, the research team used an artificial neural network (ANN), a type of machine learning algorithm, trained on a massive dataset of 13,926 borehole records—the largest such dataset applied to AI-driven geotechnical hazard assessment in Japan. The model accurately predicted soil types and N-values—key indicators of soil density and strength—at unsampled locations. By integrating these predictions into a liquefaction potential index, the team generated a detailed hazard map that identifies high-risk zones with far greater precision than previous efforts. The new map reveals previously undetected vulnerabilities in areas like Koto Ward, where soft soils and reclaimed land increase liquefaction risk. The ANN-based framework outperformed conventional methods and other machine learning models in both accuracy and ability to model complex, nonlinear relationships in geotechnical data. The resulting map provides city planners and engineers with a powerful tool for making informed decisions about construction, foundation design, and soil improvement projects. Professor Inazumi emphasizes that the model is not only effective for Tokyo but also scalable and adaptable for other megacities facing similar geohazard challenges. The framework can be integrated into geographic information systems for interactive, real-time risk visualization, supporting public education and disaster preparedness initiatives. This research marks a significant step forward in using artificial intelligence to enhance urban resilience. By combining advanced machine learning with comprehensive geotechnical data, the study sets a new benchmark for proactive earthquake risk management, contributing to safer, more sustainable urban development in high-risk regions worldwide.