A framework for Machine Learning- based Fall from Height Prediction in Construction Industry

Abstract

The construction industry is undeniably one of the most hazardous sectors, where
workers face a multitude of risks daily. Among these risks, falls from height (FFH)
stand out as a significant concern, accounting for a substantial proportion of fatal
and nonfatal injuries. Over the years, with the advent of advanced technologies
and data analytics, there has been a growing interest in leveraging Machine
Learning (ML) and artificial intelligence (AI) techniques to enhance fall risk
assessment and prevention. This paper provides a comprehensive, concept-centric
literature review of FFH, exploring its evolution, diverse models, the use of
machine learning and artificial intelligence techniques for better assessment and
prevention as well as extensive applications. This paper presents a framework for
an explainable machine learning-based model for proactive FFH prediction of in
construction sites. The framework leverages the predictive power of random forest
classifier, a robust ensemble learning method, along with the interpretability
offered by the Local Interpretable Model-agnostic Explanations (LIME)
framework. It also critically addresses key challenges such as lack of transparency
in the use of machine learning models in FFH predictions and its consequent effect
of limiting trust among users. By evaluating the evolution and current state of FFH
research, this paper reviewed the significant trends, uncovers existing gaps, and
suggests potential direction for future work. This research work, therefore aims to
deepen the understanding of this crucial domain in the construction industry that
is receiving traction and disturbing publicity.