Proactive approaches designed to prevent incidents before they occur are essential for achieving effective safety management. Emerging as an important component of proactive safety management, leading indicators are used to assess and control safety performance. With the aim of reducing the number or severity of worksite accidents, methods capable of predicting future safety performance using leading safety indicators have been developed. However, these methods have been developed for a specific set of leading indicators. This has substantially limited their application in practice, as leading indicators with the greatest impact on safety performance vary considerably between organizations and projects. An approach for predicting accident occurrence on construction sites that can be applied to any combination of leading indicators is proposed to address these limitations. Data used to develop the proposed approach were collected by a construction company from eight construction projects over a period of two years. Feature selection techniques were used to filter the original factors into the most critical subset, which were then used as inputs. Various supervised learning algorithms, namely support vector machine (SVM), logistic regression, and random forest, were then tested to determine which algorithm(s) yielded the highest prediction accuracy. The results demonstrate that the proposed procedure can be used for early recognition of potentially hazardous project characteristics and site conditions regardless of the number or type of leading indicators available within an organization. Research in this area is expected to facilitate the implementation of targeted safety management controls and to improve safety performance.