Enhancing identification accuracy: Novel fractional activation function as a versatile tool for improved model performance

Kumar, Meshach and Mehta, Utkal V. and Cirrincione, Giansalvo (2025) Enhancing identification accuracy: Novel fractional activation function as a versatile tool for improved model performance. Engineering Applications of Artificial Intelligence, 158 (Part B). pp. 1-15. ISSN 0952-1976

Text - Published Version Restricted to Registered users only Download (2MB) | Request a copy

Abstract

This study introduces and evaluates the novel Riemann–Liouville (RL) conformable fractional derivative-based Adaptable-Shifted-Fractional-Rectified-Linear-Unit (
) activation function within multi-layer perceptron models, showcasing its superior performance across three distinct nonlinear benchmark systems.
enhances identification accuracy by integrating a robust bias shift via the input standard deviation and extending the fractional order to
, offering greater flexibility than existing fractional ReLU variants. It demonstrates faster convergence and improved ability to capture complex nonlinear patterns. The proposed function is adaptable across multiple machine learning domains, providing a versatile tool for advancing neural network design.

Item Type:	Journal Article
Uncontrolled Keywords:	Neural networks System identification Multi-layer perceptron Activation functions Fractional Neural Networks
Subjects:	T Technology > TK Electrical engineering. Electronics Nuclear engineering > Robotics and Automation
Divisions:	Faculty of Science, Technology and Environment (FSTE) > School of Engineering and Physics
Depositing User:	Utkal Mehta
Date Deposited:	06 Aug 2025 22:01
Last Modified:	06 Aug 2025 22:01
URI:	https://repository.usp.ac.fj/id/eprint/15086

Actions (login required)

View Item