Developing a Decision Model for Enhancing the Safety of CNG Stations Using Data Mining and Safety Indicators Classification

Abstract

Abstract This study addresses the need to expand Compressed Natural Gas (CNG) filling stations regarding the increasing popularity of dual-fuel vehicles. The primary challenge in this regard is ensuring these stations’ safety and implementing effective safety measures. To this end, a decision model was developed using data-mining techniques. The data needed for this purpose included 57 CNG stations in Markazi, Tehran, Isfahan, and Khuzestan provinces (Iran). Then, a comprehensive model was formulated using the safety indicators extracted from relevant literature. The data were analyzed using classification and prediction algorithms, i.e., Naive Bayes and Apriori, respectively. Naive Bayes achieved an accuracy rate of 89.3% in predicting defects, outperforming other algorithms. On the other hand, classification using Naive Bayes assigned high priority to specific safety indicators, including compression systems, equipment safety, and site and traffic safety. In this study, driver safety received the lowest priority (with a mere 1% allocation), followed by employee safety (at 2%) and environmental and vehicle safety (at 3%). The Apriori algorithm revealed crisis measures required to enhance CNG station safety. These measures included environmental safety, employee safety, equipment and system maintenance, compliance with regulations, and site and traffic safety. The sensitivity analysis highlighted that employee and driver safety (65%) and equipment safety (35%) were particularly sensitive to CNG station safety, with training identified as the most impactful safety indicator.