Data Science in Construction

Groups I work with:

Problem

Construction is one of the most dangerous industries in the United States.

• 7.2% of the workforce

• 20% of all occupational fatalities

• Nonfatal occupational injuries occur at 5 times the national rate

• Injuries incur an estimated average annual cost of $13 billion

Construction jobsites are “fractured workplaces” consisting of multiple employers/contractors/subcontractors

Safety Prequalification

Job owners use Safety Prequalification surveys to evaluate subcontractors

Previous work

Estimating ASSOCIATIONS between safety indicators and safety outcomes using classical model building techniques (Zero-Inflated Poisson models)

Current projects

• Machine learning methods to predict safety performance from historical safety prequalification data

• Cluster subcontractors on organizational safety values enabling owners to identify companies sharing similar safety values

• Developing safety prequal dashboards

Current/Future work

• Methodological issues around updating/retraining ML models

• Web scraping projects regarding smaller construction companies

• Ethical issue around historical disparities propagating through models biasing results

• Sentiment analysis of data collected from a safety communication RCT

THANK YOU