With the remarkable advancements in artificial intelligence and machine learning, intrusion detection systems (IDS) in computer networks have undergone a fundamental transformation. These technologies provide an effective tool for enhancing cybersecurity by analyzing vast amounts of data and identifying anomalous patterns. However, detecting rare and imbalanced attacks within high-volume network traffic remains a significant challenge, where traditional methods often perform poorly. In this study, a multilayer perceptron (MLP) neural network was employed to detect intrusions in the CICIDS2017 dataset, which contains over 5.9 million samples of normal traffic and diverse attack types, including DoS Hulk, PortScan, DDoS, and Web Attack – XSS. After removing irrelevant features, 78 selected features remained, and the Synthetic Minority Over-sampling Technique (SMOTE) was applied to address data imbalance. The dataset was split into training (64%), validation (16%), and test (20%) subsets. The proposed model consisted of a 10-dimensional input layer, a hidden layer with 128 neurons, and incorporated Batch Normalization and a 0.5 Dropout mechanism, and it was trained over 40 epochs. Evaluation results demonstrated that the model achieved an overall accuracy of 88%, a weighted average F1-score of 0.90, and an AUC of 0.98, indicating satisfactory performance. Nevertheless, the model exhibited poor detection of rare attacks such as Web Attack – XSS, reflecting mild overfitting and residual data imbalance. These findings highlight the necessity of employing more advanced balancing techniques and deep learning architectures to improve the detection of infrequent attacks.