Intrusion Detection Systems (IDS) are crucial in safeguarding network security against increasingly sophisticated threats. In this study, we propose a Transformer-based intrusion detection model to enhance attack recognition performance. The UNSW-NB15 dataset is utilized for model training and evaluation. The data preprocessing pipeline includes handling missing values, encoding categorical features, normalizing numerical features, and splitting stratified training and testing sets. The Transformer model has three layers, leveraging self-attention mechanisms to capture relationships between network features. Experimental results demonstrate that the model achieves an accuracy of 98.26% and an F1-score of 95.80%, outperforming traditional methods such as Random Forest. Notably, despite exhibiting a higher false alarm rate, the model significantly reduces the number of undetected attacks. The Transformer demonstrates superior performance and strong potential for real-time cybersecurity applications compared to previous studies. Future research directions include enhancing the model’s interpretability, optimizing its deployment in resource-constrained environments, and extending its capability to detect zero-day attacks.