The second Strategic Highway Research Program (SHRP 2) Naturalistic Driving Study (NDS) offers a unique glimpse into alcohol-impaired driving through the inclusion of an alcohol sensor within the standard instrumentation package. This research effort developed and evaluated an alcohol-detection algorithm using the sensor through two approaches: an experimental in-vehicle testing regimen and an examination of a subset of SHRP 2 NDS trips. For the experimental in-vehicle testing, a sedan was instrumented with two SHRP 2 alcohol sensors. During 50 15-minute trials, controlled levels of alcohol were introduced into the cabin and alcohol sensor readings were recorded. The sensitivity of the algorithm at detecting alcohol presence during these trips was 96.6% and the specificity was 100%. A data set was created using SHRP 2 trips that were visually reduced via manual video coding for the presence of imbibed and unimbibed alcohol (n = 659). It provided insight into many unimbibed alcohol sources that can lead to misclassification as imbibed alcohol. Confusion matrices were conducted on the data set and a subset comprising trips with moderately impaired drivers and “normal” driving trips. The results indicated a sensitivity ranging from 92.2% to 93.7% and a specificity ranging from 36.9% to 100%. The large variance in specificity is due to one of the data sets intentionally oversampling “positive” cases. The results indicate that an alcohol-detection algorithm can be a useful tool for identifying potential alcohol-impaired trips in the SHRP 2 database. However, trained data reductionists should also be used to make final impairment determinations due to the large number of unimbibed alcohol sources that can affect the sensor. (Author/publisher)
Abstract