Decision support model for prioritizing railway level crossings for safety improvements: Application of the adaptive neuro-fuzzy system

• Published in: Expert systems with applications 2013-05, Vol.40 (6), p.2208-2223
• Main Authors: CIROVIC, Goran, PAMUCAR, Dragan
• Format: Article
• Language: English
• Subjects: Applied sciences; Artificial intelligence; Buildings. Public works; Computer science; control theory; systems; Connectionism. Neural networks; Decision theory. Utility theory; Exact sciences and technology; Fuzzy logic; Fuzzy set theory; Ground, air and sea transportation, marine construction; Level crossings; Neuro-fuzzy model; Operational research and scientific management; Operational research. Management science; Railroad crossings; Railroads; Railway accidents; Railway engineering; Railway level crossings; Railways; Road transportation and traffic; Roads; Safety improvements; Transportation infrastructure
• ISSN: 0957-4174; 1873-6793
• DOI: 10.1016/j.eswa.2012.10.041

Every year, more than 400 people are killed in over 1.200 accidents at road-rail level crossings in the European Union (European Railway Agency, 2011). Together with tunnels and specific road black spots, level crossings have been identified as being a particular weak point in road infrastructure, seriously jeopardizing road safety. In the case of railway transport, level crossings can represent as much as 29% of all fatalities caused by railway operations. In Serbia there are approximately 2.350 public railway level crossings (RLC) across the country, protected either passively (64%) or by active systems (25%). Passive crossings provide only a stationary sign warning of the possibility of trains crossing. Active systems, by contrast, activate automatic warning devices (i.e., flashing lights, bells, barriers, etc.) as a train approaches. Securing a level crossing (whether it has an active or passive system of protection) is a material expenditure, and having in mind that Serbian Railways is a public company directly financed from the budget of the Republic of Serbia, it cannot be expected that all unsecured level crossings be part of a programme of securing them. The most common choice of which level crossings to secure is based on media and society pressure, and on the possible consequences of a rise in the number of traffic accidents at the level crossings. The process of selecting a level crossing where safety equipment will be installed is accompanied by a greater or lesser degree of uncertainty of the essential criteria for making a relevant decision. In order to exploit these uncertainties and ambiguities, fuzzy logic is used in this paper. Here also, modeling of the Adaptive Neuro Fuzzy Inference System (ANFIS) is presented, which supports the process of selecting which level crossings should receive an investment of safety equipment. The ANFIS model is a trained set of data which is obtained using a method of fuzzy multi-criteria decision making and fuzzy clustering techniques. 20 experts in road and rail traffic safety at railway level crossings took part in the study. The ANFIS model was trained with the experiential knowledge of these experts and tested on a selection of rail crossings in the Belgrade area regarding an investment of safety equipment. The ANFIS model was tested on 88 level crossings and a comparison was made between the data set it produced and the data set obtained on the basis of predictions made by experts.