This paper presents an approach for intrusion detection within IoT networks using ML techniques to further enhance modern Intrusion Detection System (IDS); concretely logistic regression and deep neural network (DNN). The results of the initial run are cryptically bad; this study highlights how much a lack of feature scaling is harmful, and why it’s important to balance data (you should look into preprocessing for both). After following these techniques for preprocessing, accuracy and F1-scores of the model showed a tremendous improvement, thus proving how important pre-processing becomes to make performance in each second. Although the DNN model achieved a significantly higher accuracy but relatively low F1-score illustrating problems in balancing precision and recall. This emphasis the problem of obtaining very high sensitivity while maintaining maximum precision. Additional investigation of several preprocessing techniques, such feature importance further improved the predictive performance for both models. Empirically Logistic Regression performs quite badly, with accuracy 18.37% and F1-score of just 10.87%. However, once some preprocessing steps were performed (e.g. feature scaling and data balancing), the results grow significantly to 75.18% for accuracy rates on both training/testing sets respectively, instead of a mere 58%. Research paths in the future will focus on the ensembled strategies enhancing particular model strengths, deep learning algorithm optimization for minority class predictions, boost real-time detection capabilities. In either case, being able to scale their capability from domain-to-domain and configuration-to centralize will also be crucial for real-world usability.