From raw to refined: Data preprocessing for construction machine learning (ML), deep learning (DL), and reinforcement learning (RL) models

Abstract

As the use of predictive models in construction rapidly increases, the need for preprocessing raw construction data has become more critical. This systematic review investigates data preprocessing techniques for machine learning (ML), deep learning (DL), and reinforcement learning (RL) models in the construction domain. Through a comprehensive analysis of 457 studies, the prevalence of six data types (i.e., tabular, image, video frame, time series, text, and point cloud) and their respective preprocessing methods are examined. Key findings reveal data transformation, cleaning, reduction, augmentation, and scaling as fundamental preprocessing categories, with applications varying across data types. The paper highlights knowledge gaps, including limited synthetic data adoption, lack of standardized annotation practices, absence of comprehensive preprocessing frameworks, and need for automated labeling. Furthermore, critical considerations regarding data privacy, security, sharing, and management practices are discussed. The review underscores the pivotal role of robust data preprocessing in enabling reliable predictive models.