Temporal Autoregressive Matrix Factorization for High-Dimensional Time Series Prediction of OSS

Area of Science:

• Computer Science
• Software Engineering
• Data Science

Background:

• Open-source software (OSS) is vital in modern development.
• Predicting OSS future development is essential but challenging.
• OSS behavioral data are high-dimensional time series with noise and missing values, requiring scalable models.

Purpose of the Study:

• To propose a scalable framework for predicting open-source software development.
• To address challenges posed by noisy, high-dimensional, and incomplete time series data in OSS behavioral datasets.
• To enhance the accuracy and versatility of time series prediction models for OSS development.

Main Methods:

• Developed a Temporal Autoregressive Matrix Factorization (TAMF) framework.
• Extracted trend and period features using a trend and period autoregressive model.
• Employed graph-based matrix factorization (MF) to impute missing values by leveraging time series correlations.
• Utilized a trained regression model for final predictions on target data.

Main Results:

• TAMF demonstrated good scalability for high-dimensional time series data.
• The framework achieved high prediction accuracy on real-world OSS behavioral data from GitHub.
• TAMF proved versatile across different types of time series data.

Conclusions:

• The proposed TAMF framework offers a robust solution for predicting open-source software development.
• TAMF effectively handles noisy and incomplete high-dimensional time series data.
• The framework's scalability and accuracy make it suitable for practical applications in OSS development analysis.