<![CDATA[The logistics sector faces increasing challenges related to road safety and operational efficiency due to unsafe driving behavior, high accident rates, and inconsistent delivery performance. Traditional monitoring approaches are often reactive and limited in their ability to capture complex driving patterns in real time. This study proposes an artificial intelligence–based driver behavior scoring framework that integrates in-vehicle telematics, GPS route data, and in-cab monitoring to improve safety performance and delivery efficiency in logistics operations. The research utilizes historical and real-time vehicle data collected from onboard diagnostic systems, including speed, acceleration, braking patterns, and driving duration. Three artificial intelligence models Random Forest, Long Short-Term Memory (LSTM), and a hybrid CNN–LSTM were developed and evaluated to classify risky driving behavior and predict safety-critical events. Experimental results indicate that the hybrid CNN–LSTM achieved the best performance, reaching an accuracy of 96.1% and a mean absolute error of 0.054. A three-month pilot deployment in a logistics fleet environment further demonstrated practical benefits, with average driver safety scores improving from 78.4 to 89.7 and on-time delivery rates increasing from 91.2% to 96.5%. These findings highlight the effectiveness of multimodal driver behavior analytics in simultaneously enhancing road safety and logistics performance. The proposed framework provides actionable decision-support insights for fleet managers and contributes to the advancement of AI-enabled intelligent transportation and logistics systems.]]>
