Final Thesis: Machine Learning for Predicting Inner Source Viability

Abstract: The adoption of inner source software development practices has grown substantially in recent years, yet organizations continue to lack data-driven instruments for assessing the viability of inner source initiatives prior to resource commitment. This thesis addresses that gap by designing, implementing, and evaluating a machine learning-based system for predicting inner source project viability. Following the Design Science Research Methodology, the artifact operationalizes the inner source viability framework of Hirsch and Riehle (2022) into a continuous scoring algorithm, combines repository-derived GitHub signals with user provided organizational inputs, and trains a LightGBM regressor alongside a linear regression baseline on a purpose-built synthetic dataset of 150 simulated projects. Inner source viability is forecasted via recursive multistep forecasting on unfinished projects. Results demonstrate that the system is functional in principle, while identifying real-world data acquisition as the decisive prerequisite for any production deployment.

PDF: Bachelor Thesis

Reference: Juri Kembügler. Machine Learning for Predicting Inner Source Viability. Bachelor Thesis. Friedrich-Alexander-Universität Erlangen-Nürnberg: 2026.