Building Information Modelling is becoming increasingly used for Asset Information Management in Facility Operations, where semantic and relational information are of primary importance. "Big Data" analytics tools provide new opportunities within this domain to classify and synthesize data, integrate it with the Computer-Aided Facilities Management system, and develop predictive models to assign priority and resources to address issues arising. The resulting information integrated into building information models provides a powerful tool for facilities management teams to prioritize and streamline operations and maintenance tasks.

This paper presents the development, comparison, and application of two supervised machine learning models to classify and evaluate maintenance requests generated both from within the maintenance team and occupant complaints. Three algorithms: Term Frequency (TF), Term Frequency-Inverse Category Frequency (TF-ICF), and Random Forest are used to analyse the text of the maintenance request description and assign problem types to each. Approximately 150,000 historical maintenance requests were used for model development and the models have overall prediction accuracies of 69%, 70%, and 90% for problem type prediction, and 92.7% for a two-stage prediction of cause code.