Offline Learning for Acknowledgment Optimization in Massive <scp>LoRaWAN</scp> Networks

ABSTRACT LoRaWAN faces critical limitations in large‐scale deployments due to spectrum contention, duty‐cycle restrictions, and energy constraints. These challenges impact the reliability and responsiveness of bidirectional communication particularly when conventional acknowledgment mechanisms are used. This paper proposes an optimized acknowledgment control strategy based on an offline‐trained logistic regression model. By inferring the necessity of downlink acknowledgments based on real network context, the method selectively suppresses redundant ACKs to reduce congestion and energy waste. The solution is fully compliant with LoRaWAN standards and introduces minimal computational overhead. Simulation results demonstrate improved delivery performance, enhanced downlink responsiveness, and increased energy efficiency under varying network densities highlighting the model's effectiveness in supporting massive IoT deployments.