Year: 2026
Type: article
Abstract: Software is an integral asset for any organization, but it needs to be properly maintained and reliable. Software change-proneness prediction is a key part of an efficient software development and subsequent software maintenance phase. In this work, we present a new approach in which the CNN model is combined with oversampling (OS) to enhance the prediction performance and generalisation. We extracted and transformed data from multiple sources, including source code metrics and change logs, to ensure consistency and relevance. The model automatically learns patterns from the data and hyperparameter optimization was conducted to boost predictive performance. Results reveal that the proposed model consistently outperforms traditional models by achieving 94.58% accuracy, an F1-score of 0.9166, and a ROC-AUC of 98.05%, compared with SVM, Random Forest, RNNs, and GNNs. The proposed model predict the probability of future modification in a software module using random oversampling with CNN. Cross validation shows good consistency and stability of performance when applied to different data sets using performance metrics. However, the issues of data quality, codebase size and model interpretability were still addressed, indicating that there is room for improvement. The findings of this study indicate that DL techniques, in particular CNN models with an OS, have much potential for providing actionable insights into software change management by reducing the cost and enhancing resource allocation in software engineering.
Source: Scientific Reports
Authors: Ruchika Malhotra, Jyoti Patidar
Institution: Delhi Technological University
Language: English
Subfield: Information Systems
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $2,690
OpenAlex ID: W7203751592
FWCI:
Cites:
Cited by:
Related to: