Advanced Salp Swarm Algorithm Based Hyper-Parameter Optimization for Cell-Level Traffic Prediction Model

Deep learning (DL) has been widely used for cell-level traffic prediction and achieved state-of-the-art prediction accuracy in recent years. Though hyper-parameters seriously impact the DL-based prediction models' performance, finding the best hyper-parameters for various prediction models is a significant challenge in the fifth generation (5G) and beyond mobile networks because optimizing each model's hyper-parameters manually with expert experience or with the exhaustively searching method is highly time and computational resource consuming. This work formulates the hyper-parameter optimization problem (HPO) related to every cell-level traffic prediction task into a combinatorial programming (CP) problem to address this issue. To solve it, we propose a salp swarm algorithm with chaotic mapping and adaptive learning (SSA-CMAL). Our numerical results demonstrate that compared with the benchmarks, the proposed algorithm has a breakneck convergence speed and can provide better hyper-parameters for the cell-level traffic prediction models to obtain higher prediction accuracy.