Scholarlink

<p>Safety improvement of roads requires accurate identification of road crash hotspots. Hotspot identification (HSID) models available in literature use crash records and/or road, traffic and traffic regulatory information for identifying potentially unsafe locations. The available Bayesian Hotspot Identification (HSID) models cannot be directly applied to urban road networks. This is because they require crash prediction models and crash modification factors, which are predominantly developed for highway environments. Thus, the hotspot analysis of urban roads relies on crash based HSID models, and these models use only crash occurrence and severity information and do not consider randomness in crash severity outcomes. The present work proposes a novel crash-based HSID model framework using Bayes Theorem, named <i>Expected Property Damage Only</i> (<i>ExPDO</i>) model, which considers all dimensions of crash information available and also incorporates randomness in crash severity outcomes. Randomness in crash severity outcomes was incorporated by estimating expected severity outcomes given the crash had occurred in a given situation, based on information available in crash reports. Performance of HSID models depend on choice of segmentation method. Spatial Hotspot Analysis with <i>Getis-OrdGi*</i> statistic and <i>Moran’s I</i> has been observed to be most appropriate for hotspot identification for urban road networks. Hotspot analysis performance of proposed <i>ExPDO</i> and conventional <i>PDO</i> models with <i>Getis-OrdGi*</i> statistic and <i>Moran’s I.</i> were compared using four-year crash records (2015–2018) from road network of Patna, India. Site Consistency, Method Consistency and False Identification tests were used for evaluation and <i>ExPDO</i> model was observed to more reliable compared to conventional <i>PDO</i> model. ExPDO improved SCT from 93.54 to 96.77, MCT from 0 to 1 (99% confidence), and reduced false positives compared to PDO, emphasizing its overall effectiveness.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Development of a novel crash based hotspot identification model considering multiple dimensions of crash reports

  • 期刊论文

  • 发表日期：

    2026年1月10日

查看全文
  • Santanu Barman,
  • Ranja Bandyopadhyaya

摘要

Safety improvement of roads requires accurate identification of road crash hotspots. Hotspot identification (HSID) models available in literature use crash records and/or road, traffic and traffic regulatory information for identifying potentially unsafe locations. The available Bayesian Hotspot Identification (HSID) models cannot be directly applied to urban road networks. This is because they require crash prediction models and crash modification factors, which are predominantly developed for highway environments. Thus, the hotspot analysis of urban roads relies on crash based HSID models, and these models use only crash occurrence and severity information and do not consider randomness in crash severity outcomes. The present work proposes a novel crash-based HSID model framework using Bayes Theorem, named Expected Property Damage Only (ExPDO) model, which considers all dimensions of crash information available and also incorporates randomness in crash severity outcomes. Randomness in crash severity outcomes was incorporated by estimating expected severity outcomes given the crash had occurred in a given situation, based on information available in crash reports. Performance of HSID models depend on choice of segmentation method. Spatial Hotspot Analysis with Getis-OrdGi* statistic and Moran’s I has been observed to be most appropriate for hotspot identification for urban road networks. Hotspot analysis performance of proposed ExPDO and conventional PDO models with Getis-OrdGi* statistic and Moran’s I. were compared using four-year crash records (2015–2018) from road network of Patna, India. Site Consistency, Method Consistency and False Identification tests were used for evaluation and ExPDO model was observed to more reliable compared to conventional PDO model. ExPDO improved SCT from 93.54 to 96.77, MCT from 0 to 1 (99% confidence), and reduced false positives compared to PDO, emphasizing its overall effectiveness.