A hybrid intrusion detection framework integrating multi-layer perceptron, SMOTE, and non-IID federated learning with explainable AI

This study introduces a novel intrusion detection system (IDS) framework that combines Federated Learning (FL), Multi-Layer Perceptron (MLP) with Synthetic Minority Over-sampling Technique (SMOTE), and Explainable Artificial Intelligence (XAI) techniques. The primary objective is to create a privacy-preserving IDS capable of addressing the increasing cyber threats while maintaining data confidentiality in distributed settings.

The framework optimizes MLP with SMOTE to handle class imbalance in the CICIDS2017 dataset and employs SHAP, PCA, and t-SNE for feature interpretability. The proposed approach ensures leakage-free performance by fitting SMOTE and feature scaling exclusively on the training partition. Under non-IID conditions, the federated framework maintains a weighted F1-score of 0.9167, demonstrating its effectiveness in a decentralized learning environment.

The framework offers an optimal balance between lightweight local threat detection, enhanced minority class recognition, and privacy-preserving decentralized learning for edge computing.

Brief written by urgent.news from Scientific Reports's own syndicated text. Machine-written — it may contain errors, so check the original before relying on it.