A Machine Learning Approach to Benchmarking Social Debt

IEEE Access, vol. 14, pp. 113135–113149

Abstract

Software development is a highly complex socio-technical process where interpersonal inefficiencies, termed social debt (SD), cause significant implicit costs and reduced productivity. Current research primarily utilizes statistical models relying on strictly additive effects, which may oversimplify the non-linear interactions between team traits and situational stressors. This study addresses this gap by introducing an analytical framework based on interpretable machine learning (ML) to quantify and benchmark SD. We employed a multi-stage methodology: variables identified in prior qualitative research were first refined through a preliminary quantitative selection ( $n=42$ ), followed by a quantitative survey of additional 101 professionals. Using this data, we developed a Polynomial ElasticNet model to detect complex patterns, operationalizing SD through the proxy of perceived additional work effort. Our findings demonstrate that the Polynomial ElasticNet approach ( $R^{2} = 0.236$ , MAE = 2.33 hours) partially addresses the limitations of traditional linear modeling by effectively handling multicollinearity and capturing non-linear interaction effects that are typically lost in additive regressions. The model identifies individual stress as a significant isolated predictor. Furthermore, the results suggest critical non-linear synergies: for example, the data indicate that high-quality project planning can mitigate the negative impacts of perceived skill imbalances. This research contributes a diagnostic benchmarking framework that enables practitioners to compare team profiles against model-derived baselines. By bridging behavioral science variables with data-driven ML analysis, this work provides a scalable foundation for managing the human factors that influence technical success.

Authors 5

1. Philipp M. Zähl Aachen Institute for Digitalization Aachen

   FH Aachen

   Affiliation as printed

   Institute for Digitalization Aachen, FH Aachen University of Applied Sciences

2. Leonie Schulte Aachen Institute for Digitalization Aachen

   FH Aachen

   Affiliation as printed

   Institute for Digitalization Aachen, FH Aachen University of Applied Sciences

3. Sabrina Cornelia Eimler

   Ruhr West University of Applied Sciences

   Affiliation as printed

   Institute of Computer Science, Ruhr West University of Applied Sciences

4. Mathias Eggert Aachen Institute for Digitalization Aachen

   FH Aachen

   Affiliation as printed

   Institute for Digitalization Aachen, FH Aachen University of Applied Sciences

5. Martin Wolf Aachen Institute for Digitalization Aachen

   FH Aachen

   Affiliation as printed

   Institute for Digitalization Aachen, FH Aachen University of Applied Sciences