Replication Package: Interpretable Sparse Autoencoders for Software Metric Reduction in Defect Prediction

Read full text Summarise with VIRAT
Authors: Dalila Amara
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

This replication package contains the Python source code used in the experiments reported in the paper: Interpretable Sparse Autoencoders for Software Metric Reduction in Defect Prediction The framework consists of three main stages: Stage 1 Unsupervised Sparse Autoencoder (SAE) training: SAE models are trained independently on each software project without using defect labels for metric contribution analysis. Stage 2 Bottleneck analysis and stability-aware metric selection: The learned bottleneck representations are analyzed to quantify the contribution of individual software metrics. Contribution frequencies are aggregated across projects and independent training replications, while inter-replication variability is considered to identify a stable set of influential metrics. Stage 3 Defect prediction validation: The resulting reduced metric set is evaluated using Support Vector Machine (SVM) and Random Forest (RF) classifiers and compared with the Full Set and conventional metric-reduction baselines.

Fields of Study & Subjects

Software Engineering Research Software Engineering Techniques and Practices Software Reliability and Analysis Research

Publication details

Journal / source
Zenodo (CERN European Organization for Nuclear Research)
Publication date
28 March 2026
Language
EN