Network intrusion detection systems (NIDS) is very important in ensuring the security of computer networks by detecting and preventing unauthorized access and malicious activities. Given the escalating complexity and variety of cyber threats, conventional rule-based IDS frequently prove inadequate. This paper undertakes an assessment of both shallow and DNN for NID, aiming to assess their performance, Competence in detecting various types of intrusions. The study utilizes a comprehensive dataset containing real-world network traffic data, including It encompasses the identification of both regular and malicious activities. Various SNN architectures, including multilayer perceptron (MLP) and convolutional neural network (CNN), are explored in this context. and recurrent neural network (RNN), are implemented and compared against DNN architectures, such as deep MLP, deep CNN, and long short-term memory (LSTM). The evaluation is conducted based on key Evaluative measures such as accuracy and precision are considered as performance metrics, recall, F1-score, and area under the receiver operating characteristic curve (AUC-ROC). Additionally, the computational efficiency and resource requirements of each network architecture are considered. The results demonstrate that DNN generally outperform SNN In relation to network intrusion detection, the assessment is based on overall accuracy and detection rates. Specifically, the deep CNN and LSTM models show superior performance, achieving high accuracy and low false-positive rates. However, SNN can still provide acceptable performance in certain scenarios where computational resources are limited. Furthermore, the paper discusses the drawbacks associated with employing neural networks for network intrusion detection, including the need for large-scale labelled datasets, model interpretability, and potential adversarial attacks. It also offers perspectives on potential future avenues for enhancing the effectiveness of neural network-based IDS.