Previous methods for levee breach closure planning rely largely on on-site decisions made by technical personnel

lacking systematic support from historical cases and resulting in low efficiency and low accuracy in emergency response. This study develops a collaborative weight-based case recommendation system for levee breach closure

integrating historical case data to lay a basis for decision-making in disaster mitigation. It focuses on the cases of levee breach closures that occurred in China in recent years

and examines the key features in selecting the closure schemes in engineering practice. By adopting correlation analysis with co-occurrence level modification

we complete the missing values in levee breach characteristics

and construct a database for the closure cases. And

we use label encoding to unify semantic and numerical features and a random forest algorithm to assess their objective weights for selecting breach closure schemes. A computational procedure for determining the subjective similarity weights of contingency schemes is also designed. The results indicate the road condition is a dominant factor in selecting breach closure schemes. Our random forest model

focusing on this feature and combined with collaborative weights

achieves an average accuracy improvement of 10% - 25% in similarity matching calculations

compared to other feature-focused combinations. And its selected closure schemes are effective and applicable to on-site emergency rescue operations.