Abstract: Existing methods for levee breach closure planning predominantly rely on on-site decisions by technical personnel, lacking systematic support from historical cases, which compromises both efficiency and accuracy in emergency response. To address this limitation, this study proposes a collaborative weight-based case recommendation system for levee breach closure, integrating historical case data to provide a scientific foundation for decision-making in disaster mitigation. This paper collects domestic cases of levee breach closures from recent years and summarizes the key features in the selection of closure plans in engineering practice. By employing correlation analysis modified by co-occurrence levels, missing values in breach characteristics are completed, and a database of levee breach closure cases is established. Furthermore, label encoding is introduced to unify semantic and numerical features, and random forest algorithm is used to assess their objective weights in the selection of breach closure plans. A computational process for determining the subjective similarity weights of contingency plans is also established. The results indicate that the feature of road conditions plays a dominant role in the selection of breach closure plans. The random forest model, when focused on this feature and combined with collaborative weights, achieves an average accuracy improvement of 10-25% in the similarity matching model. The retrieved plans provide valuable reference points during emergency rescue operations on-site.