Driving Safety Performance Evaluation Method for Heavy Vehicle Drivers Based on Super-efficiency Data Envelopment Analysis

To solve the problems of safety performance evaluation for heavy vehicle drivers in the intelligent networked environment, such as the diversity of indicators, reliability of models, integrity of evaluation, and retroactivity of results, this study proposes a driving safety performance evaluation framework for heavy vehicle drivers based on super-efficiency Data Envelopment Analysis (DEA). This framework includes methods for extracting driving behavior indicators, DEA with zero values, and driving-safety performance improvement plans based on the Efficiency Frontier Analysis (EFA). Based on the characteristics of naturalistic driving data of heavy vehicles in networked environments, six trip-level dangerous driving behavior indicators were extracted as model input indicators: speeding, harsh acceleration and braking (representing aggressive driving); yawning, cell phone use, smoking (representing distracted and fatigued driving). Trip duration and mileage are the output indicators that represent driving risk exposure. Each driver was regarded as an independent Decision-Making Unit (DMU). Three evaluation models were constructed to assess driving performance based on the dimensions of aggressive driving, distracted and fatigued driving, and overall driving risk. Furthermore, the EFA was deployed to identify inefficient drivers and quantify the driving behavior indicators that need to be improved. The framework was applied to 34 drivers in a fleet of heavy vehicles in Nanjing, China, and driver performance was evaluated using three consecutive months of networked data. The results show that the framework can accurately calculate driving performance scores with significant behavioral differences between performance levels. Speeding and yawning are key factors affecting driving performance evaluation results. Personalized driving performance improvement schemes have been proposed for inefficient drivers. Compared with existing evaluation methods, the proposed framework thoroughly mines naturalistic driving data characteristics, comprehensively measures driving styles, and multidimensionally evaluates performance to guide drivers toward safer behaviors and lower risks. The research results have broad applications in enhancing online safety supervision and risk control capabilities and establishing personalized performance evaluation systems for heavy vehicle drivers in transport enterprises and fleets. © 2023 Xi'an Highway University. All rights reserved.