Validation of network screening methods on rural highways

Abstract

Road safety in rural areas poses challenges due to lower traffic volume, random crash occurrences, and limited access to data and technical expertise for roads managed by local agencies. Consequently, current network screening methodologies and related safety analysis tools become less effective in addressing these features. While numerous network screening techniques have been developed and proposed in literature, each with its own set of strengths and limitations, there has been relatively less research dedicated to evaluating their performance. This thesis evaluates the efficacy of different network screening methods for identifying safety improvement sites on rural highways. It comprehensively assesses three methods: the Global Risk Scoring Method (GRS), the Empirical Bayes (EB) Prediction Method, and the Crash Risk Index (CRI) method. These methods are specifically designed for rural highways, considering the challenges faced by local agencies. Their performance is compared against well-established methods using rural highways in Oregon as a case study. The research employs a robust methodological framework, including spearman's rank correlation coefficient, true positive identification, and root mean square error, among other metrics, to assess the effectiveness of each method in network screening. The validation analysis was conducted separately for all methods, revealing that although the GRS method is less predictive, it outperforms the well- established empirical Bayes (EB) method in network screening. This makes it a viable option for local agencies with limited data resources. Similarly, the CRI method either outperformed or showed comparable results to the EB method in network screening across various evaluation metrics considered. The EB prediction method demonstrates comparable performance to the EB method assessed using crash frequency. The method consistency test within the network screening methods for two periods suggests that the EB prediction method excels in identifying the same hotspots across different time frames.