GOAT method: Green Orthogonal Array Tuning method

This paper is a natural extension of our previous work and introduces an eco-efficient and integrated approach to hyper-parameter optimization (HPO) using Taguchi’s orthogonal array tuning method (OATM), which forms the basis for our GOAT (Green Orthogonal Array Tuning) method across leading models...

Full description

Saved in:
Bibliographic Details
Published in:Alexandria engineering journal Vol. 133; pp. 13 - 41
Main Authors: Ranković, Nevena, Ranković, Dragica
Format: Journal Article
Language:English
Published: Elsevier B.V 01.12.2025
Elsevier
Subjects:
Eco-efficiency
Hyper-parameter optimization
Orthogonal array tuning method
Robust design of experiment
Robust design of experiment
Orthogonal array tuning method
Hyper-parameter optimization
Eco-efficiency
ISSN:1110-0168
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:This paper is a natural extension of our previous work and introduces an eco-efficient and integrated approach to hyper-parameter optimization (HPO) using Taguchi’s orthogonal array tuning method (OATM), which forms the basis for our GOAT (Green Orthogonal Array Tuning) method across leading models in Machine Learning (ML), Deep Learning (DL), and Graph Neural Networks (GNNs): XGBoost, LightGBM, CatBoost, LSTM, GRU, GGNN, and GGSNN. Taguchi’s method requires fewer than 10 experiments and just 11 s of running time for all models, demonstrating its efficiency. GGSNN emerges as the best-performing model overall. A comprehensive case study on software estimation, using 46 publicly available datasets, highlights the method’s ability to reduce time and energy consumption while improving accuracy, promoting sustainable practices and high-impact real-world applications. •Hyperparameter tuning transforms good models into state-of-the-art performers.•Evaluation shows that GOAT method outperforms existing frameworks.•GOAT method identified the top 5 hyper-parameters for the best models in machine learning, deep learning, and graph neural networks.•GOAT method requires fewer than 10 experiments and 11 s, making it eco-efficient.
ISSN:1110-0168
DOI:10.1016/j.aej.2025.10.044