Deep learning models have achieved so much attraction toward classification and prediction tasks with an impressive performance in the application of speech recognition, natural language processing, computer vision, and games playing applications with human brain-like precision. Deep learning models also proved their significance with impressive performance for cybersecurity applications, but these algorithms themselves are vulnerable to adversarial attacks. An adversary who has knowledge about the architecture of Deep Learning-based security tool can generate imperceptible noise to evade the decision. In this work, the vulnerability analysis of Deep Neural Network (DNN) architecture against evasion attacks for the implementation of intrusion detection systems is presented. We assess the impact of evasion attacks generated using the Fast Gradient Sign Method (FGSM) over the DNN model trained using two different datasets. The experimental results show the vulnerability of deep learning model against evasion attacks. It is possible to evade the detection of attack traffic flow as benign traffic flow by the intrusion detection system attacked by adversarial evasion attack.