Open-source software (OSS) has become a foundational component of modern software ecosystems. Early prediction of OSS sustainability is crucial for identifying potential risks and supporting timely intervention.Existing approaches formulate OSS sustainability prediction task as a binary classification problem using Deep Learning (DL) models.However, their black-box nature obscures the decision process and undermines the credibility of the predictions.To address this limitation, we propose SustainSR, an interpretable symbolic regression–based method for OSS sustainability prediction.SustainSR reformulates prediction task as a function-learning problem and uses Genetic Programming (GP) algorithm to learn explicit sustainability expressions.We further apply Taylor expansion to quantify feature importance and pairwise interactions.The quantified results facilitate early risk identification and actionable intervention guidance.We evaluate SustainSR on Apache and GitHub projects from three dimensions: fitting performance, predictive performance, and interpretability.Results show that SustainSR significantly outperforms most baselines, it improves AUC by 1%-4% and F1avg by 3%-7% over the best-performing baseline.Case studies further demonstrate that SustainSR offers actionable sustainability insights to support risk identification, project selection, and community governance.Overall, SustainSR provides an effective and interpretable solution for OSS sustainability analysis, bridging the gap between predictive performance and interpretability.

@inproceedings{paperbot12289,
  title = {SustainSR: Interpretable OSS Sustainability Prediction via Symbolic Regression},
  author = {Yang Deng and Mustafa Misir and Jinfu Chen and Jifeng Xuan},
  booktitle = {ASE},
  year = {2026},
  doi = {10.1145/3832783.3834366}
}