Software-defined network (SDN) is relatively a new approach for managing computer networks. In SDN, network administrator manages the network easily through software without deeper involvement in the network at its lower layers. For SDN monitoring and performance analysis, it becomes important to forecast the network traffic. In the proposed work, a systematic method to predict the future congestions in the network has been introduced. It applies a Machine Learning technique to fulfill number of relative performance attributes. These attributes are reliable input to Long Short-Term Memory (LSTM) neural network. The objective of this neural network is to predict congestions and activate load balancer in a timely manner. It is not feasible to extract numerical features with absolute values and use them as income with the neural network for machine learning and deep learning algorithms, especially LSTM neural networks. The best way is to provide pre-processed and easy-to-analyze income data for Artificial Intelligence (AI) techniques. The neural network therefore has clearer information no matter how much the network’s conditions change so that the network does not care how much the numerical values change as its income become within a specified range. The simulation results depict the ability of LSTM to adapt the behavior of the network performance activating load balancing through attributed features with an accuracy of up to 92% before reaching the throttling state.