PP-FEO: Privacy-Preserving Federated Ensemble Optimization for Intrusion Detection

Abstract

This repository contains the source code and reproducibility materials associated with the Privacy-Preserving Federated Ensemble Optimization (PP-FEO) framework for intrusion detection in distributed Cyber-Physical System (CPS) environments. The implementation includes data preprocessing and feature harmonization, the defined 42-feature harmonized representation, non-IID client data construction, Random Forest (RF) and Gradient Boosting Machine (GBM) classifiers, federated LSTM training using Federated Averaging (FedAvg), L2-norm clipping and Gaussian perturbation of complete client LSTM updates, and reinforcement learning (RL)-based ensemble-weight optimization. The repository also includes materials for ablation analysis, reward-function analysis, federated and Q-learning convergence analysis, statistical evaluation, and classification-performance assessment using accuracy, precision, recall, F1-score, ROC-AUC, and normalized confusion matrices. The Gaussian client-update perturbation implemented in this repository is an update-level perturbation mechanism and is not presented as record-level differential privacy, DP-SGD, Gaussian Differential Privacy, or a formal cumulative (ε, δ)-differential privacy guarantee. The repository is intended to support reproducibility and transparent evaluation of the computational methodology reported in the associated manuscript.