This study aims to enhance facility management (FM) in small buildings by integrating diverse computerized maintenance management system (CMMS) data sets and applying machine learning (ML) to improve text classification for maintenance work orders, while benchmarking single-building performance.
Two maintenance data sets, one from a single office building (2,596 work orders) and another from public campus facilities (117,173 work orders), were integrated using feature engineering, natural language processing (NLP) and a support vector machine (SVM) classifier to categorize mechanical, electrical and plumbing (MEP) issues. Model performance was validated through cross-validation and benchmark analysis of facility conditions.
Integrating the campus data set with the office data set improved prediction performance, achieving 85% accuracy and an 85% F1 score, with a 19% higher accuracy and 20% higher F1 score for plumbing classifications. Benchmark analysis against campus facilities enabled diagnosing performance gaps in the office building, supporting data-driven decisions.
The framework enables facility managers to automate data management, prioritize maintenance tasks and make cost-effective retrofitting decisions, enhancing efficiency in small buildings.
Prior research has overlooked small data sets from single or small-to-medium-sized buildings due to data scarcity. This study offers a novel framework for integrating large, open-source CMMS data sets with small data sets, advancing automated FM in resource-constrained settings.
