Benchmarking Hyperparameter Optimization Techniques for ML-Based Small-Signal Modeling of GaN HEMTs

Rauan Ibragim, Saddam Husain, Galymzhan Nauryzbayev, Sultangali Arzykulov, Mohammad Hashmi

IEEE Access, 2026

Full text Open access: Gold Metadata (CSL-JSON)
DOI 10.1109/access.2026.3743002
Data confidence 85 high
How this score is calculated

It is built from checkable properties of the record itself: whether the article has a DOI, how many independent sources confirmed its data, whether the required fields are filled in, and how recent the latest confirmation is. Disagreement between sources and a missing identifier lower it. A value of 80 or above means the record was published without further checking; 60 to 79 means it is published at reduced confidence, with a manual check advisable. A record verified by an editor by hand is never pushed below 90 by a recalculation. This scores our data about the article, not the article itself.

Abstract

In recent years, Machine Learning (ML) has emerged as an effective approach for small-signal modeling of Gallium Nitride High Electron Mobility Transistors (GaN HEMTs). Although various ML models and hyperparameter optimization techniques have been investigated, a systematic comparison of their performance and computational cost under a consistent benchmarking framework remains limited. This article presents a comprehensive comparative assessment of tree-based algorithms and Artificial Neural Networks (ANNs), integrated with eight distinct hyperparameter optimization methods, to establish an accurate and efficient framework for GaN HEMT S-parameter modeling. The optimization and modeling frameworks were developed in the Python programming environment, encompassing search-based, probabilistic, and metaheuristic strategies. Specifically, the tree-based models were optimized using Scikit-learn’s Random Search (RS), Grid Search (GS), and Halving Grid Search (HGS); Optuna’s Tree-structured Parzen Estimator (TPE) and Gaussian Process (GP) samplers; Particle Swarm Optimization (PSO), Grey Wolf Optimizer (GWO), and PyGAD’s Genetic Algorithm (GA). For ANN based models, hyperparameter optimization was performed using RS, PSO, TPE, and GWO. Model performance was evaluated using mean squared error, the coefficient of determination, and computational cost. The results show that the optimized models achieve comparable prediction accuracy across most optimization techniques, despite differences in their underlying mechanisms. However, a clear distinction emerges when prediction accuracy and computational cost are jointly considered, with RS and GP providing the most favorable trade-off.