Deep Learning-Empowered Framework for Performance Prediction in IEEE 802.11ah-Based Networks

With the rapid advancement of information technologies, supporting massive Internet of Things (IoT) connections remains a significant challenge for wireless communication systems. The IEEE 802.11ah standard introduces the restricted access window (RAW) mechanism to reduce station (STA) collisions by grouping STAs into separate RAW groups. In dynamic IEEE 802.11ah-based environments, accurately predicting RAW performance is particularly challenging. To address this issue, this paper proposes a bidirectional long short-term memory (BiLSTM) deep learning framework for predicting network performance. Simulation results demonstrate that the Bi-LSTM model achieves superior prediction accuracy in terms of normalized mean squared error. Compared to the multilayer perceptron model, the Bi-LSTM reduces throughput error by $\mathbf{5 4. 5 \%}$, delay prediction error by $42 \%$, and collision probability prediction error by $65.8 \%$.