Secure and Efficient Data Processing using Hybrid Federated Learning with Extreme Gradient Boosting
Srimaan Yarram, Muhamed Husseyn, P. Nagarathna, Raja Shekhara Chary Marrelli, G Bindu · 2025
Over the past few years, secure and efficient data processing has become crucial due to the exponential growth in data transfer and increasing complexity of digital systems. However, Federated Learning (FL) combined with Long Short-Term Memory (LSTM) based secure and efficient data processing faced slow convergence for non-sequential data, vulnerability to gradient based privacy attacks and high computational overhead. Hence, to address these limitations, FL combined with Extreme Gradient Boosting (XGBoost) is proposed for secure and efficient data processing. Initially, various network traffic attacks data is collected from the Telemetry Operating system and Network traffic- Internet of Things (TON-IoT) dataset. After that, the collected data is preprocessed using K-Nearest Neighbor (KNN) imputation to handle missing values, One Hot Encoding (OHE) for encoding categorical features to numerical binary vectors and min max normalization to normalize the continuous numerical features. After that, Mutual Information (MI) is applied on the normalize features for selection of most relevant features. Furthermore, XGBoost is employed to classify the network traffic data into normal or attack classes. Lastly, Homomorphic Encryption and Differential Privacy (DP) are employed for secure and efficient data transmission