An Enhanced Security Framework for IoT Devices through Federated Learning (Published)
The increasing deployment of Internet of Things (IoT) devices across diverse environments has introduced significant security challenges, particularly due to the distributed nature of IoT networks and the vast amount of sensitive data they generate. This research addresses the pressing issue of enhancing IoT security by proposing a decentralized and privacy-preserving approach that integrates federated learning models for intrusion detection. The proposed system leverages TensorFlow, Keras, and TensorFlow Federated libraries, implemented in the Python programming language, to train local models across multiple IoT clients. Each client learns from its own partition of the KDDCup 1999 dataset, a widely recognized benchmark in network intrusion detection. The system was evaluated across key performance metrics including accuracy, detection rate, and classification reliability. Experimental results demonstrated a consistent improvement in model accuracy from 93% and detection rate from 92% over 40 epochs. The distribution of detected attack types such as DDoS, phishing, malware, and ransomware further showcased the system’s practical applicability in heterogeneous IoT environments. This study confirms that federated learning is a viable approach to securing IoT systems, as it supports accurate and scalable threat detection while upholding data privacy. The model not only enhances trust and security but also demonstrates adaptability across various IoT scenarios with constrained computational resources. These scores clearly illustrated the advantage of the federated model in speed and responsiveness.
Keywords: federated learning, intrusion detection, iot security, privacy preservation, ransomware
Enhancing Cybersecurity with Machine Learning: Development and Evaluation of Intrusion Detection Systems (Published)
The widespread adoption of digital networks and information systems has transformed modern society, but it has also led to a surge in sophisticated cyber threats such as malware, phishing, denial-of-service (DoS) attacks, ransomware, and advanced persistent threats (APTs). Traditional rule-based security systems are increasingly ineffective against these evolving threats, often failing to detect novel attack patterns, leading to false positives, missed detections, and delayed responses. This study aimed to address these challenges by applying machine learning algorithms to improve the accuracy and efficiency of cyber-attack detection. Using the UNSW-NB15 dataset, which contains 175,341 training and 82,332 testing records representing both benign and malicious network traffic with 49 relevant features, the research applied synthetic minority over-sampling technique (SMOTE) to balance the dataset and principal component analysis (PCA) to reduce feature dimensionality by retaining up to 95% of data variance. Five machine learning models Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Artificial Neural Network (ANN), Decision Tree, and Random Forest were trained and evaluated using metrics such as accuracy, precision, recall, and F1 score.The results demonstrated that KNN achieved the highest accuracy of 94.69%, with balanced precision (95.31%), recall (93.96%), and F1 score (94.63%), showing robust classification of both attack and non-attack instances. Random Forest and ANN also showed strong performances with accuracies of 92.81% and 95%, respectively, highlighting their effectiveness in handling complex cybersecurity data. SVM and Decision Tree had slightly lower accuracies of 90.88% and 92.22%. These findings confirm the value of machine learning, especially KNN and ensemble methods, for real-world intrusion detection. Regular model retraining is essential to address emerging attack patterns and maintain effective cybersecurity defenses.
Keywords: Feature Selection, cyber security, cyber threats, intrusion detection, machine learning