Construction industry has one of the highest accident and fatality rates among other major industries, with more than 60,000 fatal accidents each year worldwide. Falling from height is one of the leading causes of fatalities and injuries in construction. Passive protection devices (e.g., safety net) have been used to minimize the impact of falling from height for ages, while proactive warning systems appear recently to alert the workers when they are at risks of falling. To provide appropriate warnings to the worker but not to distract them due to the false alarm, the falling risk needs to be carefully evaluated. In this paper, the authors introduced algorithms for falling risk prediction and evaluated their performance. Injuries records during 2005 to 2015 were extracted from the OSHA database and 1161 intact falling-related record were used in this study. K-Modes, RBF network and Decision Trees are chosen to build three risk prediction models, and the performance of those three proposed models were evaluated using the OSHA injuries record data. The results indicate that the DT-based falling risk prediction model has the best performance of 75% and the top three critical factors of falling event’s severity are distance from the ground, worker’s occupation and the source of the falling. The delivered severity prediction model provides the foundation of more accurate real time risk evaluation for workers at height.
Falling from height; machine learning; serverity of injury; risk prediction