Abstract:
Numerous industries have undergone radical change as a result of the Internet of Things' (IoT)
innovative solutions for event-driven applications like real-time monitoring and emergency
detection. Packet loss, congestion, and excessive latency are major challenges that affect the
reliability and performance of IoT networks, especially those with limited resources. These
problems require efficient congestion control techniques that are tailored to resource-constrained
scenarios.
A novel congestion control method is proposed in this work, which combines Random Early
Detection (RED) with machine learning-based binary classifiers. The binary classifiers predict
congestion states and dynamically modify RED thresholds based on important network metrics,
such as queue length, packet latency, and packet loss rate. The RED algorithm is dynamically
adaptive, datasets are prepared, and models are trained using techniques like Random Forest and
Support Vector Machine. The goal of this approach is to minimize traffic surges.
The findings demonstrate that the suggested method works noticeably better than conventional
RED systems. Congestion estimations become more accurate and flexible due to the Random
Forest classifier's high 95% accuracy rate. Through dynamic threshold changes, the system
dramatically reduced packet loss and delays in simulations, demonstrating its usefulness for
resource-constrained IoT networks supporting event-driven applications. Additionally, the
suggested approach demonstrates durability and scalability in a variety of IoT applications. This
paper advances the field of IoT by addressing the shortcomings of static models in dynamic IoT
situations by offering an efficient and adaptable congestion control strategy. Future work will
include additional optimization for different traffic patterns and network configurations as well as
real-world implementation.