An integrated machine learning model for aircraft components rare failure prognostics with log-based dataset

Predictive maintenance is increasingly advancing into the aerospace industry, and it comes with diverse prognostic health management solutions. This type of maintenance can unlock several benefits for aerospace organizations. Such as preventing unexpected equipment downtime and improving service quality. In developing data-driven predictive modelling, one of the challenges that cause model performance degradation is the data-imbalanced distribution. The extreme data imbalanced problem arises when the distribution of the classes present in the datasets is not uniform. Such that the total number of instances in a class far outnumber those of the other classes. Extremely skew data distribution can lead to irregular patterns and trends, which affects the learning of temporal features. This paper proposes a hybrid machine learning approach that blends natural language processing techniques and ensemble learning for predicting extremely rare aircraft component failure. The proposed approach is tested using a real aircraft central maintenance system log-based dataset. The dataset is characterized by extremely rare occurrences of known unscheduled component replacements. The results suggest that the proposed approach outperformed the existing imbalanced and ensemble learning methods in terms of precision, recall, and f1-score. The proposed approach is approximately 10% better than the synthetic minority oversampling technique. It was also found that by searching for patterns in the minority class exclusively, the class imbalance problem could be overcome. Hence, the model classification performance is improved.