The Internet of Things (IoT) devices have introduced a wide range of security vulnerabilities, necessitating robust threat detection mechanisms. This paper analyzes the role machine learning models in detecting IoT threats, utilizing the TON-IoT dataset. The proposed framework consists of a sequence of processes, including data preprocessing, statistical analysis, feature selection, classification, and evaluation. In the data preprocessing phase, duplicate entries and irrelevant features are removed to ensure data quality. The visualization process relies on correlation analysis to explore relationships between features. For feature selection, the SelectKBest method is utilized to detect the most informative features to reduce dimensionality and improving the performance of the proposed framework. The classification phase utilizes RandomFforest classifier, chosen for its effectiveness in dealing with complex datasets. To optimize model performance, Grid Search is used for hyperparameter tuning, to ensure the selection of best parameter suitable for classification process. In the experimental results, Multiple classifiers—Random Forest, Gradient Boosting, Decision Trees, and Support Vector Machines (SVM)—were used, where the proposed framework achieves the highest accuracy, up to 100% on the entire dataset and 98.79% after balancing using downsampling, outperforming existing models and current studies.