Intrusion Detection Systems are of paramount importance in network security. However, in real-world scenarios, they always suffer from the challenge of class imbalance, which is dominated by normal traffic. This paper presents a novel approach to enhancing the performance of IDS by proposing a hybrid of the Random Under sampling technique with the univariate feature selection technique, SelectKBest, for handling both problems of class imbalance and high dimensionality. This model was hence tried on the Bot-IoT dataset, which is a real-world IoT network traffic representation. The SVM classifier, which has been trained with the resampled and feature-selected data, showcased 95% balanced accuracy for both normal and malicious traffic detection. The combination of RUS and SelectKBest, apart from reducing overfitting, ensured the retention of the most relevant features and thereby made the IDS model robust. It can practically enhance the performance of IDS in an imbalanced and high-dimensional dataset by providing a balanced, efficient, and precise detecting mechanism.