Software Defined Networking (SDN) is a rising network systemwhich offers theconcept of separationdata plane from the control plane.Main focus of the control plane is to govern traffic while Data Plane is used in traffic forwarding the based on the control plane's decisions. Also, the SDN unifies the control plane, allowing numerous data plane’scomponents to be controlled by a controllerapplication. The SDN’s control plane exercises direct control over the components and the state of dataplane. Unfortunately, the SDN network architecture is complicated and poorly built to support fine-grained security policies. This is due to lack of the tight assimilationof the control and data planes. Another SDN's core challenges is figuring out how to efficiently handle network packet anomalies because SDN is logically centralized, an anomalous control plane attack can bring the entire network system down. The existing SDN anomaly detection methods, on the other hand, are known for their low accuracy and overall performance. They do not, however, facilitate gradual learning. As a result, improving and optimizing existing methods and establishing high-precision as well as real-time classification models for abnormal flows is critical. Also the flow detection mechanism must be able to support giga-bit level networks in real-time. The goal of this study is to develop such efficient anomaly detection algorithm for SDN networks.The work demonstrates this by utilizingDeep Neural Networks combined with Transductive learningfor identifying anomalous flows. An attempt is made to further improve the learning capabilities of deep neural networks andsimultaneously reducing the false positives. This is achieved by processing the input flows using transductive confidence machines and then enabling deep neural networks tomake better decisions during the learning process. Also a new transduction measure named skewness is added in this work to work alongside the exiting transduction measures. In comparison to previous techniques, the proposed TL-DNNalgorithm gives considerably reduced False Positives and greater Accuracy based on experimental findings on the KDDDataset. The work also demonstrated the superiority of the suggested method in terms of FPR, TPR, MSE, and Accuracy through various simulations.

Software Defined Network (SDN), Transductive Confidence Machines, Deep learning, Densenet