CatBoost‐SMO‐Based Congestion Prediction and Avoidance Using Rap Protocol in Active Distribution Networks

Internet of Things (IoT) devices communicate with the internet to exchange data, requiring an efficient network. IoT monitors events via sensor nodes, sending the collected data to an IoT gateway. The network's rapid growth has led to congestion, causing delays, packet loss, reduced throughput, and poor energy efficiency. Predicting congestion is crucial for minimizing data loss. IoT networks generate heterogeneous, nonlinear data due to the variety of devices and complex interactions. Nonlinear datasets provide challenges for traditional machine learning algorithms, and overfitting can lead to misclassification of real network traffic. To address these constraints, a machine learning and protocol‐based approach is used to predict congestion and congestion avoidance in active distribution networks. Initially, input is collected from the congestion detection evaluation dataset and preprocessed using the missing completely at random (MCAR) method to replace missing values. Ordered quantile normalization (ORQN) is used to standardize values in the input dataset. Next, the dataset's dimensionality gets reduced using uniform manifold approximation and projection (UMAP). After that, starling murmuration optimization (SMO)–based categorical boosting (CatBoost) is used for congestion prediction. Lastly, the rate aware protocol (RAP) is applied to mitigate congestion within the IoT network. A simulation study shows that the proposed technique attains 97.44% accuracy, 97.12% recall, and 97.74% precision. Analytical simulations show that the RAP protocol achieves better performance, lower energy consumption, longer network lifetime, higher throughput, and lower delay than existing protocols. Thus, this proposed approach is the best choice for congestion prediction and data transmission for avoiding congestion in active distribution networks.