Load Balancing Optimization through Multi-Label Classification of Network Traffic Using Deep Learning in Distributed Computing Systems

This study discusses the application of a deep learning-based multi-label classification method integrated with an adaptive load balancing mechanism in a distributed computing system. The main objective of this study is to improve the efficiency, stability, and responsiveness of the system in handling complex and dynamic network traffic. The simulation process begins with traffic data processing, feature extraction using a Recurrent Neural Network (RNN) with a Long Short-Term Memory (LSTM) architecture, and the implementation of a multi-label classification framework. The classification results are then used as input for the adaptive load balancing mechanism, which is further optimized through integration with Deep Reinforcement Learning (DRL). Performance evaluation shows that the multi-label RNN model is able to achieve an Area Under the Curve (AUC) value close to 0.95 on training and validation data, indicating good generalization ability. However, the confusion matrix reveals that there are still quite high classification errors, especially in Video Streaming and Web Traffic, whereas performance on VoIP is relatively more stable. The implementation of DRL is proven effective with a consistently increasing reward trend, indicating the agent's ability to adapt to system conditions. Furthermore, the results of throughput and latency measurements show significant improvements after the implementation of adaptive load balancing. Average throughput increased from 85–115 requests/s to 120–150 requests/s, whereas latency decreased from 95–120 ms to 65–95 ms.