This research focuses on detecting Protocol-Based Attacks (PBAs) in IoT networks using machine learning models, specifically Logistic Regression and Support Vector Machines (SVM). Protocol-based attacks exploit vulnerabilities in network protocols, making their detection critical for securing IoT environments. The methodology involves preprocessing a comprehensive dataset consisting of both benign and malicious traffic, with a focus on protocol-specific features such as packet size, IP addresses, and flow behaviors. Feature selection through Recursive Feature Elimination (RFE) refines the model by identifying the most relevant attributes, reducing complexity without compromising accuracy. The Logistic Regression model, known for its efficiency in binary classification, is rigorously trained and evaluated using key metrics including accuracy, precision, recall, and F1-score. The evaluation process demonstrates the superior performance of Logistic Regression in minimizing false positives and accurately classifying threats. This research highlights the significance of using Logistic Regression for detecting PBAs in IoT networks, offering a scalable and reliable solution for enhancing IoT security frameworks against emerging cyber threats.