PREDICTION OF PROBABILITY OF DROWNING INCIDENTS AT BEACHES BY AI MODEL CONSIDERING FEATURE IMPORTANCE AND ITS GENERALIZABILITY

Abstract

 In Japan, 2,000 to 3,000 drowning accidents occur every season at major bathing beaches. To prevent drowning accidents, beachgoers themselves need to be aware of risks of the bathing beach and avoid them. As a way to do this, bathing beaches provide daily risk levels regarding swimming conditions to beachgoers using beach safety flags in three colors. We have created an AI model that can predict the probability of drowning accidents with high accuracy and developed a new method for objectively judging swimming conditions based on the prediction results. On the other hand, the actual application of this method to be used at beaches requires the continuous collection of 53 features and the versatility of the AI model. In this study, we investigated a creation of the AI model that can predict the probability of drowning accidents with a small number of features and its versatility. As a result, even the AI model using 10 features with high importance was able to predict with high accuracy, and its versatility was also demonstrated.