Collaborative Research (Georgia Tech)
Research Motivation
Facility maintenance data sets have not been actively employed because of missing data and data inconsistency. This research attempts to resolve the issues (i.e., missing data and data inconsistency) by proposing a systematic approach that leverages machine learning-based text classification algorithms.
Key Idea
This research specifically utilizes four different classification algorithms (i.e., Support Vector Machine (SVM), Multi-Layer Perceptron, Random Forest, and Naïve Bayes) and evaluates the performance of the algorithms to identify the most appropriate prediction model. A case study is constructed with 3,632 HVAC-related maintenance requests of higher education buildings retrieved from Computerized Maintenance Management System (CMMS) software as a proof-of-concept.
Overview of the proposed methodology
Results
The results show that the best performance of the prediction model (e.g., the capability to predict missing data correctly) with the SVM achieves an 85% accuracy rate compared to the other algorithms. The outcome of this research can improve the performance/efficiency of the data-driven decision-making processes in the Facility Management field by providing the ability to predict missing data inputs more consistently. It implies the potential to apply the text classification methods in fault impact analysis from occupant comfort aspects.
Publications
Click here to access the publication.
Comments