Daily traffic accidents increase annually, causing a significant number of death and disability cases. Most of fatalities occur because of the late response to these emergency cases. The time after the traumatic injury is called the golden hour, where providing essential medical and surgical aid at that time increases the probability of saving human lives by one-third an average. Thus, the focus of this paper was to develop a system based on IoT for accident detection and classification. The system detects and classifies vehicle accidents based on severity level and reports the essential information about the accident to emergency services providers. The system consists of a microcontroller, GPS, and a group of sensors to determine different physical parameters related to vehicle motion. In addition, different types of machine learning classifiers were examined with the developed system to determine the most accurate classifier for the system. The classifiers are the Gaussian Mixture Model (GMM), Naive-Bayes Tree (NB), Decision Tree (DT), and Classification and Regression Trees (CART). The implementation of the system showed that GMM and CART models were better in terms of precision and recall. It was also shown that the severity of accidents depends mainly on the g-force value and fire occurrence.