Enhancing MANET Security Through Long Short-Term Memory-Based Trust Prediction in Location-Aided Routing Protocols

Résumé fourni par la source

Mobile Ad-Hoc Networks (MANETs) face significant security threats due to their decentralized and dynamic topology, making them susceptible to malicious attacks. Traditional routing protocols struggle to maintain security while ensuring network performance. To address this, we propose LSTMT-LAR, an LSTM-based trust prediction mechanism integrated with Location-Aided Routing (LAR) to enhance MANET security. LSTMT-LAR utilizes a 13-feature behavioral model to assess node trustworthiness in real-time, enabling proactive detection of malicious nodes. The protocol was evaluated across four scenarios with varying node densities (30-100 nodes) and malicious node percentages (10-90%). Results show that LSTMT-LAR maintains a high Packet Delivery Ratio (PDR) of 0.786 even with 50% malicious nodes, outperforming conventional protocols. Furthermore, it optimizes energy consumption by reducing overhead while maintaining competitive end-to-end delays. These findings validate LSTMT-LAR’s effectiveness in securing MANET communications, demonstrating its potential for practical deployment.