<scp>MoCLEO</scp> : A Multiparameter Optimization Study for Efficient Hyperparameter Tuning for Time‐Series Prediction Using Collective Support Strategy

ABSTRACT Over the years, many machine learning models have been developed to deal with various domain data, ranging from classification models and predictive models. Although the models can effectively handle various tasks, there are still some challenges associated, such as getting an effective model parameter configuration. With that in mind, researchers are constantly proposing approaches that will mitigate those challenges. In this paper, we propose MoCLEO (Multiparameter Chimp Leader Election Optimization Algorithm) a streamlined variant of the CLEO algorithm, applied to LSTM‐based time‐series prediction for hyperparameter tuning within the model, finding Parameter configurations that improve the performance of a given model (LSTM). The study focused on Multiparameter, simultaneously turning three LSTM hyperparameters: LSTM unit (U), learning rate (LR), and batch size (BS). Adopting MoCLEO, which is inspired by the concepts of chimp leader election with a collective support strategy. The algorithm has three stages: challenge, support, and updating phase. The work also includes the projection‐based fitness‐aware parameter updates, aiming for a balance, adaptive, and systematic parameter exploration. We finally conducted an experiment on three well‐known real‐world time‐series traffic datasets, PeMSD7‐M and L, METR‐LA. While the experiment result demonstrated a comparative performance, it offers the advantage of faster computational efficiency compared to other baseline approaches. The explainability analysis indicates a balanced prioritization of hyperparameters, reflecting the diverse search effort of the algorithm's internal PBU and RSU mechanisms.