<![CDATA[The rapid escalation in the sophistication of network attacks has exposed the limitations of traditional Intrusion Detection Systems (IDS). While machine learning has shown great promise in enhancing IDS performance, its success often hinges on the effectiveness of feature selection. Standard feature selection techniques, however, struggle in cybersecurity applications due to the highly imbalanced nature of network traffic datasets. In such settings, minority attack classes, though critical, are often overshadowed by majority classes, leading to reduced detection of rare intrusions. To address this challenge, we propose a hybrid feature selection framework that integrates Analysis of Variance (ANOVA) and Mutual Information (MI) with a novel class-frequency weighting mechanism. This weighting scheme adjusts the relevance score of each feature according to the distribution of classes, ensuring that features associated with rare attacks are more strongly emphasized during the selection process. We evaluate our method on the UNSW-NB15 dataset using a Support Vector Machine classifier. The results show that our approach achieves substantial gains in recall for underrepresented classes while simultaneously reducing feature dimensionality and maintaining efficiency. By improving the visibility of features tied to minority attacks, the proposed framework provides a more balanced and reliable solution for modern IDS. This contribution advances the detection of rare but impactful threats and highlights a scalable pathway for building more resilient cybersecurity defenses.]]>
