Smart home networks powered by Internet of Things (IoT) technology offer enhanced efficiency and convenience through interconnected device communication. Because of the growing number of connected devices increases the attack surface and increases susceptibility to threats like botnet attacks. Existing rule-based security methods often fail to manage the complexity of IoT environments. Machine learning (ML) offers a promising solution through adaptable, scalable techniques that address the current security approaches. Unlike rule-based methods, ML models can recognize traffic patterns dynamically, and enabling real-time identification of botnet attacks. This study addresses the gap by evaluating the potential of machine learning (ML) classifiers as robust alternatives for real-time botnet detection. ML algorithms were tested on two comprehensive NetFlow datasets, NF-ToN-IoT and NF-BoT-IoT, to assess their performance in binary classification tasks. The methodology included SMOTE for balancing class distribution and stratified K-fold cross-validation for reliable evaluation. Findings indicated that Random Forest (RF) and Decision Tree (DT) consistently outperformed other models, achieving approximately 99% accuracy, near-perfect specificity and sensitivity, and prediction times under 0.5 s. The NF-ToN-IoT dataset produced more consistent outcomes than NF-BoT-IoT, highlighting the significance of dataset characteristics in model performance. These findings emphasize the effectiveness of ML techniques, particularly RF and DT, in strengthening IoT network security by detecting and mitigating botnet threats.