The digital age has made cybersecurity even more important, necessitating the use of innovative Intrusion Detection Systems (IDS) to strengthen network defenses. This work introduces a novel deep learning-based Integrated IDS that uses long Short-Term Memory (LSTM) and CNN (Convolutional neural network). Suggested framework consists of three important phases: Pre-process, feature abstraction and selection, classification. During the pre-processing stage, various techniques such as label encoding, data normalization class imbalance are applied to standardise input data. The proposed two-tier method, features ranked by PCA and IG for retrieval is characterized by a high generalization capability across numerous datasets with different sizes during both feature and sample reordering. The designed IDS we have is only focused on classification, for which an efficient CNN-BiLSTM network based classifier has been introduced. This integration reduces architectural complexity to its minimum level, yet allows the model to express the temporal and spatial patterns within large-scale network data. To make the algorithm more efficient we use Improved Monarchy Butterfly Optimization. It leads to faster training rates which are meant for network datasets at a much larger scale than before. For the NSLKDD datasets, evaluate how effective optimized CNN and BiLSTM components are, with an emphasis on precision, accuracy and recall metrics. Compare present ML models for classification such as SVM, LSTM and CNN + LSTM. The results are to demonstrate the superiority of suggested architecture over other known models using its precision, accuracy and recall in detecting more kinds of attacks.