An improved approach to analyze accidents and promote road safety using association rule mining and multi-criteria decision analysis methods

Background: Road accidents have become a major social and health problem for being dramatically increasing day after day worldwide. Scientists are conducting their studies to define the main attributes that share the severity of road accidents. Finding a new approach to analyze road accidents is of great urgency. Data mining techniques are best fitting to discover useful information out of enormous data which are used to make proactive decisions. Methods: This paper tempts a rule-based machine learning method known as association rule mining, which can identify strong rules discovered in databases using interesting measures. Given a da-ta-set from the Lebanese territory for the years 2016-2017, the application of association rule mining, the Apriori method takes its place. However, its implementation leads to a very large number of rules. The task that is the most difficult is extracting meaningful and non-redundant rules. In order to find out the most interesting and relevant rules out of fatal rules such, ELECTRE TRI and PROME-THEE methods, the most significant methods of decision making, Multi-Criteria Decision Analysis (MCDA) are integrated to resolve the outranking problem. The integration is presented by the use of the same set of weights and the same constant values of Indifference and Preference thresholds used in ELECTRE-TRI method to define the linear preference function needed by PROMETHEE method. Realizing the sensitivity of the final output of alternatives ranking to the changes of the input parameters of the decision-making tool, this proposed integration helps the decision makers to overcome their ambivalence between preference and indifference thresholds and to cope adequately with the issue of the uncertainty of MCDA procedures; it comes up with the complete ranking of rules. Results: The obtained ranked rules declare the most significant attributes or combinations of attributes that influence the severity of road accidents. Four main factors of fatal road accidents are pinpointed: over-speeding mainly leading up to rollover crashes, pedestrians encountering in the context, distracted driving leading to fatal road vehicle collisions with Pedestrian victims; and wet roads particularly in the case of single car accidents. Meanwhile, the importance of ELECRE-TRI and PROMETHEE and their integration in dealing with such complex phenomena and corresponding database with a large number of involved attributes have been validated. Conclusion: This paper studies the phenomenon of road accidents. Association rule mining has been applied to discover all possible relations between the various attributes. The integration of ELEC-TRE-TRI and PROMETHEE MCDA techniques aims at extracting meaningful information from the big dataset. The obtained results have shown how influencing the behavior of the driver is on the occurrence of fatal road accidents. These findings contribute to supporting decision makers to draw new design conceptions for road infrastructure and develop preventive measures that improve road safety in Lebanon. © 2020 Bentham Science Publishers.