The ever-growing rate of mobile device adoption and data usage has posed a significant challenge for service providers in sharpening celluar network performance. Predicting mobile network traffic for celluar networks has proven to be a challenge for service providers due to its highly dynamic and nonlinear nature. In this paper, we describe a framework that uses AI technologies to foresee traffic congestion in mobile networks to enable autonomous adaptive resource distribution and congestion alleviation. Our system implements ensemble learning combining LSTM networks, GBM, and SVR. To address this issue, we apply adaptive hyperparameter tuning based on evolutionary meta-heuristics. Using traffic data from urban mobile networks, our comprehensive analysis showed that our ensemble learning AI models yielded a 27% improvement in prediction accuracy with RMSE metric compared to baseline models. These models show great adaptability for real-time operation in SON frameworks, thereby enhancing resiliency, efficiency, and usercentric self-organizing mobile communication networks.
Agrawal et al. (Mon,) studied this question.