On the Prediction of Open Source Software Ecosystem Health Based on Time Series Analysis

Open Source Software (OSS) is a cornerstone of the computing field and a staple in our daily lives, with its health and sustainability being crucial for the global software infrastructure. In the field of OSS ecosystem health assessment, most existing platforms cannot predict the future health of the project, which limits the community’s grasp of the potential risks and development trends of the project. In addition, existing OSS ecosystem health prediction research faces problems such as limited data set size, low prediction accuracy, and poor model generalization ability. To address the limitations above, we first collect extensive weekly indicator data from a multitude of open source projects and achieve multi-source data integration. Then this paper proposes an OSS ecosystem health prediction method based on frequency domain convolution to fuse adjacent frequency features. Experimental results indicate that the proposed method not only improves the performance but also provides consistent prediction accuracy and stability in multiple open source projects. Finally we implement an OSS ecosystem health prediction system, which showcase the effectiveness of our proposed method in providing a reliable assessment tool for the OSS community.