As cyber-attacks become increasingly sophisticated, making sure that strong intrusion detection mechanisms exist in Industrial Control Systems (ICS) has become extremely vital. This paper presents a data-balanced machine learning framework to improve ICS security using the HIL-based Augmented ICS Security Dataset. Seven machine learning models—Logistic Regression, Decision Tree, Random Forest, K-Nearest Neighbors, Support Vector Machine, Gaussian Naïve Bayes, and XGBoost, are compared on accuracy, precision, recall, F1-score, and false negative. Overall accuracy is consistent, but the false negative rate comes down substantially by balancing data, making the system more capable of identifying severe intrusions. Time to train and test is also compared to determine computational efficiency. These findings highlight the significance of data balancing in reducing missed attacks and enhancing the dependability of intrusion detection in ICS environments.