Yang, LinhaoLinhaoYangCorbally, RobertRobertCorballyMalekjafarian, AbdollahAbdollahMalekjafarian2024-05-212024-05-212022 CERI-2022-08-26978-0-9573957-5-6http://hdl.handle.net/10197/26015The 2022 Civil Engineering Research in Ireland (CERI) and Irish Transportation Research Network (ITRN) Conference, Dublin, Ireland, 25-26th August 2022Every year thousands of incidents occur on Irish roads. These incidents can be varied in nature and severity, ranging from debris on the road to serious road-traffic-collisions. The management of incidents on motorways is of particular importance, both in terms of road-user safety and maintaining network performance. Incident management encompasses a broad range of activities, with a multi-agency response often required to ensure that an incident is managed safely and efficiently with minimal traffic disruption. When an incident occurs on the motorway network, a dynamic risk assessment must be made by response personnel to estimate the severity of the incident and the potential impact on traffic conditions. A key parameter in this assessment is the duration of the incident, which is often difficult to establish, and likely to change as the incident evolves. Making a judgement on the expected duration of an incident can be difficult, however as traffic management processes become more automated, computer algorithms and historical incident databases can be leveraged to improve real-time predictions of incident duration. This paper analyses incidents that occurred on the M50 motorway in Ireland. By comparing the predictive performance of multiple machine learning methods for different types of incidents, an integrated approach is proposed to utilise the advantages of different methods. The results show that support vector machines perform best in most cases, but in some cases a different method may need to be used. Suggestions are made for further improvements which could improve accuracy and benefit real-time motorway operations response procedures.enTraffic incidentsM50 motorwayCongestionMachine learningRegressionClassificationConference Publication2023-08-18https://creativecommons.org/licenses/by-nc-nd/3.0/ie/