Search Results

Federated Learning (FL) Has Emerged As A Promising Paradigm For Collaborative Model Training Without Directly Sharing Raw Data, Thereby Preserving Privacy Across Distributed Parties. In Horizontal Federated Learning (HFL), Participants Hold Datasets With The Same Feature Space But Different Samples, Making Feature Selection Critical For Improving Learning Efficiency, Reducing Communication Costs, And Enhancing Model Generalization. However, Existing Feature Selection Approaches In FL Often Rely On Computationally Intensive Methods Or Expose Sensitive Intermediate Information During Aggregation, Creating Scalability And Security Challenges. To Address These Issues, This Study Proposes A Secure And Lightweight Feature Selection Framework Tailored For HFL. The Framework Integrates Privacy-preserving Techniques, Such As Homomorphic Encryption And Secure Aggregation, With Low-complexity Feature Evaluation Metrics To Enable Efficient Selection Of The Most Informative Features Without Compromising Data Confidentiality. The Proposed Approach Significantly Reduces Communication Overhead, Improves Training Convergence, And Ensures Robust Privacy Guarantees Against Inference Attacks. Experimental Evaluations On Benchmark Datasets Demonstrate That Our Method Achieves Comparable Or Superior Accuracy To State-of-the-art FL-based Feature Selection Schemes While Maintaining Lower Computational And Communication Costs. This Work Contributes To Building Practical, Privacy-preserving, And Resource-efficient HFL Systems For Real-world Applications In Healthcare, Finance, And IoT Ecosystems.