<p>Bengkulu City, located along the active Sumatra Subduction Zone, faces a high risk of liquefaction due to its geological formation, which consists of saturated Holocene coastal deposits. Although previous studies have advanced local hazard mapping, limited spatial coverage and conventional site investigation methods have constrained regional-scale assessments. This study developed data-driven techniques using a deep neural network (DNN) to enhance subsoil data derived from microtremor and borehole measurements to support spatially continuous liquefaction hazard mapping. The optimised DNN model, trained on microtremor-derived shear wave velocity profiles and topographical data, enabled the generation of high-resolution site classification, peak ground acceleration and liquefaction susceptibility assessment under the 2007 Bengkulu–Mentawai earthquake scenario. Liquefaction hazards were evaluated using the factor of safety and liquefaction potential index, while ground deformations were estimated through volumetric strain modelling and lateral displacement predictions based on the lateral displacement index. Results show a sandy-dominated stratigraphy with severe liquefaction risks concentrated in the southern parts of Bengkulu City. Ground settlements were generally shallow, while lateral displacements reached up to 3.8 metres in critical zones. Integrating DNN-based subsoil profiles with empirical hazard models enhances regional liquefaction assessments, offering a scalable and efficient tool to support seismic risk mitigation and infrastructure resilience in seismically active coastal environments.</p>