Paper Title
Lightweight VS Heavyweight Machine Learning Models for Scalable IoT Intrusion Detection

Abstract
This paper examines lightweight and computationally intensive machine learning models for detecting security threats in Internet of Things (IoT) environments. Using the CICIOT2023 dataset, multiple classification techniques including Logistic Regression, Decision Trees Models, Naive Bayes, Lin- ear SVM, ensemble-based methods, and neural networks—are implemented and evaluated under a unified framework. The data is preprocessed through class balancing and feature standardization to ensure reliable model performance. Evaluation is conducted using both predictive and efficiency evaluation metrics including accuracy, precision, recall, F1-score, ROC-AUC, training time, inference latency, and model size. The results show that while complex models achieve marginally higher detection accuracy, lightweight models offer significantly lower computational overhead and faster inference. This research highlights the significance of considering both performance and efficiency when designing intrusion detection systems. The results demonstrate that lightweight models are better suited for real- time IoT applications, where resource constraints and low latency are critical. Keywords - IoT Security, Intrusion Detection System, Machine Learning, Lightweight Models, Real-Time Detection, Scalability