IoT networks face various security threats and scalability challenges, necessitating robust and scalable solutions. This involves drop attacks, replay attacks, and tamper attacks that cause threats to the integrity and performance of a network. This study introduces a trust-driven Levy-Based Moth-Flame Optimization (LB-MFO) algorithm combined with a Weight-Tuned Neural Network (NN) to strengthen the security and scalability of IoT networks. It uses the hybrid technique of trust mechanism and LB-MFO optimization, which is highly accurate and reliable for malware node detection and classification. Experimental results demonstrate the effectiveness of the proposed methodology with excellent performance metrics, achieving 98.06%, 97.89%, 98.17%, and 98.07% in terms of accuracy, sensitivity, specificity, and precision rate on a 100-node network. These results indicate that the NN-based LB-MFO-improved approach is more effective than traditional approaches in most cases, and it is a scalable mechanism that could be used to reduce the drop, replay, and tamper attacks. Such a method becomes the yardstick of safeguarding IoT ecosystems from upcoming threats while improving scalability and reliability.