Intrusion detection system (IDS) works as heart to make a network safe. Most of the IDSs are facing the problems due to high false alarm rate (FAR) which is leading for decrease in the accuracy for big data environment when it is placed in real environment. A new, reliable and hybrid IDS model is proposed for anomaly detection in a cloud virtual network. The implemented model employs hybrid Grey Wolf Optimization - Bald Eagle Search (Hybrid GWO-BES) method and Long Short Term Memory (LSTM) for increasing the detection rate (DR) and decreasing the false alarm rate (FAR) and hence, increasing the accuracy. The proposed IDS model uses hybrid GWO-BES for feature selection. The selected relevant features work as the input to a Long Short Term Memory (LSTM) for the classification. The performance of proposed GWO-BES-LSTM model has been shown on the basis of accuracy, precision, recall, F1 Score (F Score), false alarm rate and false-negative rate. The results are simulated on NSL-KDD, and CICIDS-2017 datasets for multi class classification and compared with the existing approaches. Simulated results of the GWO-BES-LSTM model achieved an increased performance in compared to other models.