Intrusion Detection Systems (IDS) are crucial for maintaining security in Internet of Things (IoT) environments. However, the high-dimensional nature of IoT network data and the presence of numerous outliers pose significant challenges. Traditional Principal Component Analysis (PCA) has shown promise in reducing data dimensionality but suffers from sensitivity to outliers due to its reliance on the arithmetic mean. To address this issue, we propose Median PCA (MedPCA), a robust variant that replaces the mean with the median, enhancing resistance to outliers. Extensive experiments conducted on the IoT23 and CICIoT2023 datasets demonstrate that Median PCA achieves superior accuracy and reduced CPU time compared to traditional PCA, making it a robust and efficient solution for enhancing IDS performance in large-scale IoT networks.