Abstract

Over the years DDoS attack are not easily noticed and preventing them has been uneasy due to the fact that the sources addresses are initiated and other method are used to hide attack sources, Physical gadgets only cannot solve DDoS attacks. Also, because of the necessity to enhance global view of the network Data Centre, network operators have also evolved from the traditional based network to Software Defined Network (SDN) because SDN gives more reliability, flexibility and a secure network environment . Even with the enormous capabilities of SDN, it is still faced with several security challenges due to its architecture complexity. SDN architecture is very vulnerable to DDoS attack. In this research work, six (6) different Deep Neural Network (DNN) models has been analyzed and from the analysis it has been seen that The Long Shortterm Memory (LSTM) out performs other five (5) algorithms with accuracy of 100%, loss of 0.000287211 which is very low compared to other algorithms considered. LSTM takes 915,002 milliseconds per Iteration which is quite high, the Convolutional Neural Networks (CNN) takes 300,000.635 milliseconds per iteration and it has 99.99% accuracy. Therefore, fusing the LSTM and CNN models into DDoS mitigation system in software defined networks will be beneficial. This study is limited to User Datagram protocol (UDP) and Transfer Control Protocol (TCP) attacks.