Home // SOFTENG 2026, The Twelfth International Conference on Advances and Trends in Software Engineering // View article

Fine-tuned Random Forest-based Software Defect Prediction via Metaheuristics

Authors:
Tadashi Dohi
Jingchi Wu
Junjun Zheng
Hiroyuki Okamura

Keywords: software defect prediction; defect-prone module; ran dom forest; hyperparameters; metaheuristics; fine tuning; search based software engineering

Abstract:
Software defect prediction is one of the most fundamental issues to identify defect-prone modules in module testing,and enables to reduce effectively the module testing cost. While several machine learning techniques are applied to the software defect prediction, it is well known that ensemble methods such as random forest are quite useful to detect the defect-prone modules with high predictive performance and low computation cost. Since the Random Forest (RF) contains the hyperparameters, which have to be carefully tuned in advance, the prediction accuracy on the defect-proneness strongly depends on the tuning results in the random forest. In this paper, we propose to apply three metaheuristic algorithms: Latin Hypercube Sampling (LHS) algorithm, Artificial Bee Colony (ABC) algorithm, and Parameter free Genetic Algorithm (PfGA), to search the optimal hyperparameters in the RF-based software defect prediction, and investigate the predictive performance of defect-prone modules. In experiments with six actual software development projects, we compare our fine-tuned RF algorithms with the existing machine learning approaches. It is empirically shown that the fine tuning via metaheuristics could provide better predictive performances on F-score in software defect prediction.

Pages: 15 to 20

Copyright: Copyright (c) IARIA, 2026

Publication date: May 24, 2026

Published in: conference

ISSN: 2519-8394

ISBN: 978-1-68558-398-9

Location: Venice, Italy

Dates: from May 24, 2026 to May 28, 2026