Transport-related incidents cause \(\tilde{1}.25\) million deaths annually worldwide, with driver drowsiness a major factor. In India, about 328,000 accidents yearly are attributed to drowsy driving. This paper presents a real-time driver drowsiness detection system using deep learning and attention mechanisms in Wireless Body Area Networks. The system compresses a heavy baseline model into a lightweight version using facial landmark key point detection. It employs ResNet50 for feature extraction, XAI techniques like GradCam, and a modified MobileNet with spatial attention. It uses transfer learning and considers multiple drowsiness indicators (head tilting, blinking, yawning) for robust detection across various conditions. Temporal factors are included to enhance prediction reliability. Experimental results show up to 98.4% accuracy, even with drivers wearing masks or glasses. This research demonstrates the potential of advanced driver assistance systems to reduce drowsy driving risks and improve road safety, hereby assisting in proactive accident care.