Validation of WAIMSS incident duration estimation model

This paper presents an effort to validate the incident duration estimation model of WAIMSS Wide Area Incident Management Support System (WAIMSS). Duration estimation model of WAIMSS predicts the incident duration based on an estimation tree which was calibrated using incident data collected in Northem Virginia. The validation process started with collection of new incident data which was conducted by video taping incident management processes in Northem Virginia and keeping detailed incident logs of actual incidents at Northem Virginia Traffic Control Center. The collected incident data was then partitioned into a number of subsets according to the structure of the original estimation tree. Due to the limited sample size, a full scale test of the distribution, mean and variance of incident durations was performed only for the root node of the estimation tree, while only mean tests were executed at all other nodes whenever a data subset was available. Further studies were also conducted on the model error and tree structure issues especially related to complex incidents incidents with multiple major discriminating incident characteristics. The statistical analyses in general, strongly supported WAIMSS estimations of incident duration distribution, mean and variance. The error analysis provided encouraging results based on the distribution of estimation errors and estimation error percentages. A major structural deficiency of the current model was also revealed. While WAIMSS duration estimation model is effective in dealing with incidents with one important characteristic, for complex incidents with more than one discriminating characteristic, only the most significant incident characteristic is used by the current model and others are simply ignored. Thus, data analysis shows a trend of underestimation f i r complex incidents. Altematives to improve this shortcoming of the current model along with several other possible improvements are also recommended.