Supporting facility maintenance decision-making under imbalanced work orders using natural language processing and interpretable machine learning

Effective facility maintenance management relies on efficient Maintenance Work Orders (MWOs) assignments, which remain heavily manual due to unstructured data. While Machine Learning (ML) techniques show promise in MWOs assignments, previous studies emphasize boosting model performance but overlook data imbalance and interpretability. To address these limitations, this study proposes a data-driven framework integrating Natural Language Processing (NLP) and interpretable machine learning techniques to enhance MWOs assignments performance. The proposed framework consists of three modules: (1) Data preparation, (2) Model development, and (3) Interpretability analysis. Validated on real-world campus MWO data, the framework demonstrated high performance, achieving a 0.98 accuracy for predicting priority, 0.92 for predicting failure type, and 0.92 for worker assignment. Interpretability analysis results indicated that integrating temporal patterns with semantic embeddings could enhance predictive performance. This study addresses the challenge of processing unstructured and imbalanced MWOs data to support data-driven decisions, thus contributes to improving facility maintenance management practices.