Traffic incidents have a serious negative impact on safety and traffic flow, and fast accurate automatic incident detection on freeways is a major theme in transportation engineering. Therefore, various types of AID (Automated Incident Detection) algorithms have been proposed for more accurate and rapid incident detection, and Artificial Neural Network models have provided significantly improved performance in terms of detection and false alarm rates. Recently, Deep Neural Networks (DNNs) has received much attention due to its excellent performance and was also used for automatic incident detection on highways. However, in learning algorithms such as Backpropagation and SVMs(Support Vector Machines), the prediction performance is known to be highly depend on the input vector characteristics. The purpose of this study is to examine whether the input detection performance of DNNs differs according to the normalization method of the input vector and to verify how sensitive it is to the method. Furthermore, the best way to normalize the input vector of the DNNs model has been proposed in order to obtain the best performance in terms of DR (Detection Rate) and FAR (False Alarm Rate) in AID (Automatic Incident Detection).