Driving behaviors analysis for public transport drivers in Kuwait: a machine learning approach to drivers safety

This research paper addresses the critical concern of evaluating driving behaviors among bus drivers in Kuwait to enhance road safety and prevent accidents. Real driving data from 73 bus drivers working in Kuwait Public Transport Company (KPTC), collected through Teltonika devices, forms the basis of the quantitative analysis. The OPTICS (Ordering Points to Identify the Clustering Structure) algorithm and Expectation–Maximization (EM) clustering were employed using Gaussian Mixture Models (EM-GMM) to classify drivers into distinct behavioral categories. Correlation analyses were then conducted to pinpoint factors influencing risky driving. It was revealed that over speeding is the predominant contributor, accounting for 84.89% of unsafe behaviors. Predictive modeling is undertaken using Gradient Boosted Trees (GBT) and discriminant analysis, with GBT emerging as the most effective, achieving the highest accuracy. Risk indices for each driver cluster are calculated, showing that 28% of drivers exhibit unsafe practices. The probability of accidents for drivers with hazardous tendencies was determined to be 0.772, while the general likelihood of accidents among bus drivers in Kuwait is calculated at 0.318. Surprisingly, no significant correlation is found between age and driving behavior, highlighting the influence of factors such as psychological conditions, fatigue, weather, and road conditions on driving conduct. The findings contribute valuable insights for developing targeted interventions to mitigate risky driving behaviors and enhance overall road safety in the region.