This study evaluates the performance of Kolmogorov-Arnold Networks in anomalous traffic detection, a key task in development of Intrusion Detection Systems. Their ability to achieve strong results with fewer resources and explainability makes them particularly suitable for real-world cybersecurity applications. We compared Kolmogorov-Arnold Networks with Multilayer Perceptrons, Gradient Boosting, and Random Forests using the NSL-KDD dataset. One of the key advantages of Kolmogorov-Arnold Networks is their inherent explainability. After evaluation, we extract an elementary function-based expression from the trained Kolmogorov-Arnold Network, demonstrating its potential for feature relevance analysis. This capability aids in feature selection and problem characterization. The results show that Gradient Boosting achieves the highest performance across all metrics. Although Kolmogorov-Arnold Networks are outperformed by Gradient Boosting, they outperform Multilayer Perceptrons, offering superior effectiveness with a smaller neural network architecture and enhanced explainability. Among the two configurations of hidden layers for Kolmogorov-Arnold Networks, the shallower one showed better results.