Network Intrusion Detection System (NIDS) is the need of the hour to handle the increasing number of network attacks. The high accuracy achieved by the existing machine learning and deep learning approaches is also accompanied by the high FPR, which reduces the overall efficiency of the intrusion detection system. This paper considers the spatial and temporal features available in the network traffic data. A hybrid model is recommended that combines the strengths of a convolutional neural network (CNN) and bidirectional long-short term memory (Bi-LSTM) neural networks to integrate the learning of spatial and temporal characteristics of the data for intrusion detection. The data imbalance problem is also handled using SMOTE algorithm. The two benchmark datasets, NSL-KDD and UNSW-NB15, are used to train and test the model. The random forest (RF), CNN, LSTM, and CNN-Bi-LSTM classifiers are used to compare with the proposed method. The HySDL-ID model illustrates high accuracy, high precision, and low false positive rate (FPR). The comparative analysis of the HySDL-ID model with existing published work is also carried out, and the findings reveal that the HySDL-ID model shows better performance.