Year: 2026
Type: article
Abstract: Centralized intrusion detection models face challenges relating to data privacy, scalability, and collaborative threat detection across distributed network environments. This study aimed to develop and evaluate a Decentralized Intrusion Detection Model (DIDM) based on Federated Learning to improve intrusion detection performance while maintaining the locality of raw network data. The study adopted the Design Science Research Methodology (DSRM). A centralized Artificial Neural Network (ANN)-based Intrusion Detection Model (CIDM) was established as the experimental baseline. Subsequently, a federated DIDM was designed and implemented using Federated Averaging (FedAvg), enabling distributed clients to train local ANN models and collaboratively update a global model without centralizing their raw training data. CICIDS2017 was used for the principal empirical evaluation.
Author: Kennedy Omondi Okeyo
Institution: Kabarak University
Language: English
Subfield: Computer Networks and Communications
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W7228874750
