Predicting Abandonment of Open Source Software Projects with An Integrated Feature Framework

TL;DR

This work tackles OSS project abandonment prediction by defining a precise abandonment criterion that combines explicit archival status with semantic declarations in documentation, enabling high-quality labeling of a large longitudinal GitHub dataset (115,466 repos with 57,733 abandonments). It introduces an integrated, multi-perspective feature framework that includes user-centric, maintainer-centric, and project-evolution signals, and demonstrates superior predictive power using an Accelerated Failure Time model with a C-index of 0.846. The approach emphasizes interpretability through SHAP analyses and ablation studies, and validates practical utility with a GBSA classifier deployed in the openEuler ecosystem for proactive risk screening. Overall, the method offers reproducible data, scalable features, and actionable insights to support dependency management, program comprehension, and governance in large OSS ecosystems.

Abstract

Open Source Software (OSS) is a cornerstone of contemporary software development, yet the increasing prevalence of OSS project abandonment threatens global software supply chains. Although previous research has explored abandonment prediction methods, these methods often demonstrate unsatisfactory predictive performance, further plagued by imprecise abandonment discrimination, limited interpretability, and a lack of large, generalizable datasets. In this work, we address these challenges by reliably detecting OSS project abandonment through a dual approach: explicit archival status and rigorous semantic analysis of project documentation or description. Leveraging a precise and scalable labeling pipeline, we curate a comprehensive longitudinal dataset of 115,466 GitHub repositories, encompassing 57,733 confirmed abandonment repositories, enriched with detailed, timeline-based behavioral features. Building on this foundation, we introduce an integrated, multi-perspective feature framework for abandonment prediction, capturing user-centric, maintainer-centric, and project evolution features. Survival analysis using an AFT model yields a high C-index of 0.846, substantially outperforming models confined to surface features. Further, feature ablation and SHAP analyses confirm both the predictive power and interpretability of our approach. We further demonstrate practical deployment of a GBSA classifier for package risk in openEuler. By unifying precise labeling, multi-perspective features, and interpretable modeling, our work provides reproducible, scalable, and practitioner-oriented support for understanding and managing abandonment risk in large OSS ecosystems. Our tool not only predicts abandonment but also enhances program comprehension by providing actionable insights into the health and sustainability of OSS projects.

Figures (5)

• Figure 1: Methodology Overview.
• Figure 2: Lifespan distribution of abandoned repositories.
• Figure 3: Illustration of the user-repository bipartite graph.
• Figure 4: Feature importance across categories.
• Figure 5: Time-series feature dynamics for Electronic WeChat.