All Times ET
Keywords: random forest, security incidents, Shiny visualizations, forecasts, knowledge management
Reduction of security, safety, and other operational incidents is a priority for many organizations. This study demonstrates the value of incorporating data-driven analyses and knowledge management (KM) into ongoing operational activities to reduce such incidents for a large research & development institution. In our approach, a data analysis process based on machine learning evaluations is integrated with a larger decision-making framework to support selection of organizations for organizational assessment activities, which focus on targeted training to reduce incident risk. Based on performance accuracy, a random forest model with down sampling was selected to forecast an organization’s likelihood of incidents. The model output is visualized by a Shiny Applet to support operational decision making. Visualization and analysis are incorporated into a continuous integration and deployment process that ensures changes made to the code are successfully built and tested prior to publishing to the production pipeline. Since the initial implementation of the algorithm, multiple updates have been made to extend the utility of the processing pipeline to other security activities, including ingestion of additional datasets and development of a R package. These activities reflect the evolutionary process associated with KM to ensure alignment of data-driven tools with changing organizational needs.