In recent years, supply chain (SC) disruptions caused by production stoppages at bottleneck firms have been frequent due to disasters. To develop a business continuity plan (BCP) to prevent SC disruptions, it is necessary to identify the bottleneck firms. In this study, we developed a total of 7 models for extracting bottleneck firms using machine learning methods based on the centrality and geographic features of firms and transaction networks obtained from actual data. The results show that the bottleneck firms tend to be characterized by large distances between firms downstream, diverse industries. Finally, we validated the model and the accuracy of the model was high. Our method is expected to contribute to the improvement of resilience of the entire SC, such as BCP and early recovery of SCs after disasters.