Machine Learning Assessment of Lifeguard Perception of Rip-Risks

Beach user's safety is always the number one concern for lifeguards. To ensure their safety, a regularly updated flag system is often present at many popular beaches. The purpose of this study is to identify the environmental controls lifeguards use to determine flag colour and to create a model capable of interpreting real-time data to output an accurate classification. To accomplish this, a machine learning approach, in the form of both a decision tree and random forest, is implemented and trained using data collected by experienced lifeguards from Pensacola Beach, Florida, as well as the flag colour they deemed appropriate based on the beach conditions. The number of rescues on a given day is found to be partly dependent on the difference between the lifeguard and the model's decision on flag colour, suggesting an erosion of confidence in the warning system by the beach user. As an extension to this model, the implementation of a rip current classifier, using a modified version of the 'You Only Look Once' object detection algorithm, was applied to enhance the validity of the tool as it will not only take into consideration the measurements the lifeguards collect but also the visual ques they can see. Although the model will be capable of classifying the water conditions on the flag scale, the goal of the study is not to provide a substitute for the lifeguard's judgment, rather a tool to aid in the judgment process and to determine how to reduce the number of rescues and drownings when beach user perception is not consistent with the decisions made by the lifeguard.