On 28th September 2018, a very high magnitude of earthquake Mw 7.5 struck the Palu city in the Island of Sulawesi, Indonesia. The main objective of this research is to estimate the earthquake risk based on probability and hazard in Palu region using cross-correlation among the derived parameters, Silhouette clustering (SC), pure locational clustering (PLC) based on hierarchical clustering analysis (HCA), convolutional neural network (CNN) and analytical hierarchy process (AHP) techniques. There is no specific or simple way of identifying risks as the definition of risk varies with time and space. The main aim of this study is: i) to conduct the clustering analysis to identify the earthquake-prone areas, ii) to develop a CNN model for probability estimation, and iii) to estimate and compare the risk using two calculation equations (Risk A and B). Owing to its high prediction ability, the CNN model assessed the probability while SC and PLC were implemented to understand the spatial clustering, Euclidean distance among clusters, spatial relationship and cross-correlation among the estimated Mw, PGA and intensity including events depth. Finally, AHP was implemented for the vulnerability assessment. To this end, earthquake probability assessment (EPA), susceptibility to seismic amplification (SSA) and earthquake vulnerability assessment (EVA) results were employed to generate risk A, while earthquake hazard assessment (EHA), SSA and EVA were used to generate risk B. The risk maps were compared and the differences in results were obtained. This research concludes that in the case of earthquake risk assessment (ERA), results obtained in Risk B are better than the risk A. This study achieved 89.47% accuracy for EPA while for EVA a consistency ratio of 0.07. These results have important implications for future large-scale risk assessment, land use planning and hazard mitigation. © 2020 Elsevier B.V. Show More