Software Defined Networking (SDN) provides a promising networking architecture that provides a solution for traditional networks by decoupling the control plane from the data plane. With the advantage of this features, the controller can get global view of the entire network. Since, the controller acts as the brain of the network in SDN environment. However, SDN controller is mainly attacked by security problems, among such security problems the most common one is Distributed Denial of Service (DDoS) attacks which leads to exhaustion of the system resources and causes the availability of the services given by the controller. As a result, it is critical to design DDoS attack detection mechanism to mitigate the controller attack at the initial stage. In this case, one of the most promising methods to confirm SDN security is the use of deep learning technique to detect and classify the network traffic into normal and attack. We proposed a deep learning algorithm LSTM with autoencoder to detect and classify network traffic in SDN controller. In this paper we have trained our model against the recently published CICDDoS2019 dataset with a total of 190,910 dataset records were used for model evaluation. From these 152,728 datasets were used for training, 19091 datasets records were used for test set and 19091 datasets for validation. Finally, by training and testing with our model Long short-term memory (LSTM) we have achieved the highest prediction accuracy rate 98.93% and lower false positive rate of 1.07%. As compared to other benchmarking Machine Learning classification approaches Support Vector machine (SVM) and Naïve Bayes classifiers (NB), our model performance is best accuracy in protection of these network attacks.