Intrusion Detection Systems play an important role in network security. With the advent of Machine Learning and Artificial Intelligence, ML-based IDS have been gaining increasing popularity in comparison with traditional anomaly-based and signature-based ones. While bringing several advantages, these systems are vulnerable to so-called adversarial attacks, where a malicious user tampers with testing data to compromise the performance and reliability of the ML-based IDS. This research presents a novel approach to enhance the robustness of IDS against adversarial evasion attacks using a committee of Neural Network classifiers. Each model in the committee is trained on different feature subsets, specifically omitting features of high importance, which are prone to be the subject of adversarial attacks, enabling a more resilient detection mechanism. Experimental results using the CICIoT2023 dataset Mirai attack samples, as an example of malware that could use adversarial attacks to hide its presence from AI-based detectors, demonstrate the effectiveness of the proposed method.