In certain circumstances, near-miss falls can evolve into fall accidents in construction sites. Insight into near-miss falls offers an efficient way to better understand fall accidents. In this context, this paper explores potential applications of the smartphone as a data-acquisition tool to detect and identify near-miss falls on the basis of an artificial neural network (ANN). In training and evaluation experiments, a loss-of-balance (LOB) environment was artificially established by means of a balance board to simulate the scenarios in near-miss falls. Through a transition model between static and dynamic near-miss falls, the similarity between simulated and actual scenes of near-miss falls was improved. Furthermore, the feasibility of adopting ANN to correctly identify near-miss falls was verified. The results showed that the average precision, recall, and F1 score were 90.02%, 90.93%, and 90.42%, respectively, with an average error-detection rate of 16.26%. In test cases, the thresholds H20% (0.07692) and H10% (0.06061) were acquired and illustrated from the perspective of probability. This approach, which demonstrates the feasibility of integrating smartphones and ANN to measure near-miss falls, will help detect near-miss fall events and identify hazardous elements and vulnerable workers. In addition, it provides a new perspective for measuring the relationship between near-miss falls and fall accidents quantitatively, laying a solid foundation for better understanding the occurrence mechanisms of both events.