Highly imbalanced intelligent identification of civil aviation maintenance hazards based on text mining and selective ensemble modeling

Accurate identification of civil aviation maintenance hazards is of vital engineering significance for mitigating accident risks, safeguarding operational safety, and optimizing maintenance efficiency. However, the complex linguistic features and highly imbalanced distribution of civil aviation maintenance logs pose significant challenges to existing identification algorithms. To address these issues, this paper proposes a method for highly imbalanced intelligent identification of civil aviation maintenance hazards based on text mining and selective ensemble modeling. Firstly, considering the domain specificity of maintenance logs, domain-specific text preprocessing is performed, followed by the term frequency-inverse document frequency (TF-IDF) method to achieve effective feature extraction and vectorization. Subsequently, on the basis of data class rebalancing via the instance hardness threshold (IHT) sampling, an ensemble pruning strategy integrating dual correlation analysis and predictive accuracy-based filtering is designed to effectively manage the trade-off between the diversity and accuracy of the base models. Finally, an enhanced adaptive weighted fusion algorithm (EAWFA) is constructed to flexibly adjust the weights for base model fusion, thereby enhancing the precision and engineering intervenability of hazard identification. Comprehensive experiments conducted on maintenance logs from an aviation maintenance enterprise and the aviation safety reporting system (ASRS) dataset validate the effectiveness and generalization capability of the proposed method. Comparisons with existing methods demonstrate the superiority of the proposed method, indicating its promising potential for civil aviation maintenance engineering applications. • A novel selective ensemble learning framework is proposed. • An enhanced adaptive weighted fusion algorithm is designed. • Model effectiveness is validated using engineering and open-source cases.