Machine Learning Recognition Models in Construction: A Systematic Review

Abstract

Due to its growing acceptance and success in many sectors, there is a rapidly rising adoption and application of machine learning recognition models within construction. As a result of this adoption and usage surge, there is copious knowledge residing in different repositories. This surge makes it a daunting task for researchers and other stakeholders to access concise and summarised evidence of existing research showing the usage and adoption of different recognition models in construction. As a result, a systematic review of machine learning recognition models with their different applications in construction is inevitable. We leveraged PRISMA protocol and PICOC technique to retrieve 819 construction-related studies from SCOPUS. We grouped recognition models into Image Recognition, Pattern Recognition, Voice Recognition, and Natural Language Processing (NLP). Our thorough analysis and approach show that 53% of existing studies use Pattern Recognition, 42% Image Recognition, and 2% Voice Recognition. We identified that 45% of the studies focused on buildings, 31% on worker's health and safety, while 24% was on equipment detection, efficiency, and usage. We recommend that future studies leverage the textual and voice data generated from construction-related activities and studies. This will build more voice and NLP recognition models for training robots and other assistive technologies that can support construction workers to improve their safety and productivity. This study will guide researchers and other stakeholders in this field to widen their horizons on trends in recognition model application to construction, making informed decisions, and establish gaps in the literature while suggesting lasting contributions.