Intrusion Detection in IoT Environments: An Experimental Evaluation Using Multiple ToN-IoT Dataset Subsets as a Baseline for Federated Learning

Abstract

The growing adoption of Internet of Things (IoT) devices has in-
creased the need for efficient intrusion detection systems capable of identifying
cyberattacks in heterogeneous environments. This work investigates the applica-
tion of machine learning techniques for intrusion detection using data from the
ToN-IoT dataset. Experiments were conducted using Logistic Regression, Ran-
dom Forest and Multilayer Perceptron (MLP) models applied to different IoT
devices, including Weather, Modbus, Garage Door and GPS Tracker. The pro-
posed methodology includes data preprocessing, binary classification of normal
and malicious traffic, and comparative evaluation using accuracy, precision, re-
call and F1-score metrics. The study establishes an experimental baseline for
future integration with Federated Learning architectures in IoT environments.
Preliminary results demonstrate significant performance differences among the
evaluated models, highlighting the importance of model selection for intrusion
detection tasks in IoT scenarios.