| description abstract | Facility managers often need to accurately predict the annual maintenance of their building roofs to develop reliable and cost-effective maintenance plans that maximize their performance and life expectancy. This article presents the development of a novel machine learning (ML) model using XGBoost to predict maintenance costs of ethylene propylene diene monomer (EPDM) roofing systems, and compare its performance to multivariate linear regression (MLR). The two models were developed in three main phases that focused on data collection and processing, model development, and performance evaluation. The collected data include 374 historical annual maintenance records of EPDM roofs that consist of maintenance cost, age, area, opening rate, and weather data. The performance of the two developed models was evaluated using four metrics: mean absolute percentage error (MAPE), accuracy, root square mean error (RMSE), and coefficient of determination (R2). The outcome of this performance evaluation illustrates that the average accuracy of the ML model in predicting maintenance costs of EPDM roofs (88.20%) was significantly higher than the MLR model (68.30%). This highlights the original contributions of the developed ML model. The ML model has novel capabilities to provide much-needed support for facility managers to improve the accuracy of estimating the annual maintenance costs of EPDM roofs to ensure the development of reliable maintenance plans for this type of roof. | |
