As smart city infrastructures increasingly depend on interconnected digital systems, cybersecurity threats have escalated significantly, necessitating robust intrusion detection capabilities. This study investigates the impact of feature correlation analysis on machine learning-based Intrusion Detection System (IDS) performance using the UNSW-NB15 dataset for binary classification tasks. We employed three machine learning algorithms–Logistic Regression, K-Nearest Neighbors (KNN), and Linear Support Vector Machine (SVM)–trained on feature subsets selected using Pearson correlation coefficients at thresholds of 0.0, 0.3, 0.5, and 0.8. The methodology involved comprehensive data preprocessing, including null value removal, one-hot encoding for categorical variables, and Min-Max normalization. Our experimental results demonstrate that the 0.5 correlation threshold achieves optimal performance, with all three models attaining over 97% accuracy while maintaining computational efficiency. Specifically, KNN achieved 97.86% accuracy, while both Logistic Regression and Linear SVM reached 97.75% accuracy at this threshold. Lower correlation thresholds (0.3) introduced noise that degraded Linear SVM performance to 73.54%, whereas higher thresholds (0.8) retained only one feature, potentially limiting model generalizability. Cross-validation confirmed model robustness and prevented overfitting across all experimental configurations. This research establishes that systematic feature correlation analysis significantly enhances IDS effectiveness in distinguishing malicious network traffic from legitimate activities, providing a foundation for optimizing real-time cybersecurity systems in IoT-enabled smart city environments.