Overview
In this paper, a neural network model was developed to predict daily and weekly incidents in the Michigan
Department of Transportation’s (MDOT) Metro Region. Weather information from the weather station at the
Coleman A. Young Municipal Airport in Detroit was used as input data in the model. Meanwhile, the incident
data provided by the Southeast Michigan Transportation Operations Center (SEMTOC) were used for model
calibration and validation. Five years’ worth of weather and incident data from 2011 to 2016 were processed
for model calibration. The model was applied to predict the daily and weekly number of incidents and high impact
incidents for the period of September 2016 to December 2016. The prediction results show that the
model is on average 65 to 75 percent accurate in predicting the number of incidents and can capture the
trend of the incident occurrence over the weeks. Prediction errors between the predicted incidents and actual
incidents are due to either the inaccuracy in the weather forecast information or other contributing factors that
were not accounted for in the model. The errors could be minimized through more frequent updates of
weather information or incorporate other effective input variables in the model. The model can be developed
as an application tool for SEMTOC, and other first responder agencies, including Michigan State Police
(MSP), local police departments, fire departments, and towing companies, for the purpose of staff and
resource planning, and situational awareness.