Enhancing public safety by reducing traffic crashes due to human error is critical. A key strategy involves ensuring that drivers remain alert through the implementation of early safety notifications. Nonetheless, forecasting the risk of traffic crashes poses a complex challenge due to numerous factors including road conditions, traffic dynamics and weather patterns. In response to this problem, this study endeavors to develop an end-to-end traffic risk notification system, integrating IoT system with Deep Learning solution. This multi-node collaborative IoT system offers more pragmatic risk analysis by incorporating both static and dynamic factors which can contributes to a traffic crash. At the device level, strategically placed sensors capture dynamic data which is then transmitted to the decision node. Leveraging a deep learning-based model, the decision node processes both static and dynamic information to predict a crash severity risks, and the outcomes are seamlessly communicated to the display node for timely notification to drivers, fostering a safer and more responsive driving environment. The DNN model is developed via extensive training with accident history data in the state of Texas. The evaluation of our deep learning model was performed using key metrics, including accuracy, recall, precision, and F1 score. Finally, multi-node collaborative IoT system is evaluated based on prediction and communication latency.