AUTHENTICATION METHODS USING BEHAVIORAL ANALYTICS AND MACHINE LEARNING FOR INTERNET OF THINGS DEVICES

Abstract

The growing complexity of cyber threats has highlighted the limitations of traditional authentication methods, including passwords, tokens, and standard two-factor authentication (2FA). In the Internet of Things (IoT) environment, these methods are particularly vulnerable due to limited computational resources, the dynamic nature of connections, and the need for seamless user–device interaction. In response to these challenges, behavioral analytics and machine learning (ML) are gaining increasing attention as they enable the development of adaptive, continuous, and user-transparent authentication systems. This study focuses on behavioral authentication methods, including keystroke dynamics, mouse movement patterns, geolocation data, session activity, and network traffic analysis. A modular architecture is proposed that integrates both supervised and unsupervised ML algorithms, such as Support Vector Machines (SVM), Random Forest, Artificial Neural Networks (ANN), and autoencoders. Based on a combination of public and experimental datasets, extensive preprocessing and feature engineering were applied to identify the most informative behavioral characteristics of users and devices. Experimental results showed that the Random Forest model achieved the highest accuracy (96%) and F1-score (0.94), while the deployed prototype system provided fast response times (0.6 s) and a low false positive rate (0.1%) in a real-time web environment. These findings confirm the practical applicability of behavioral authentication methods for IoT, where classical approaches are often ineffective. At the same time, several key implementation challenges were identified: the need for large volumes of training data, ensuring the privacy of behavioral patterns, integration into heterogeneous IoT ecosystems, and maintaining a balance between performance and accuracy. Promising directions for further research include optimizing algorithms for resource-constrained devices and applying federated learning to minimize the risks of data leakage. Thus, behavioral analytics combined with ML forms a new paradigm of authentication, capable of providing a high level of information security in the context of the rapid expansion of IoT.