Early detection of tipping points in Software Engineering - A Cross-Domain Perspective on Software System Instability using Early Warning Signs

Sammanfattning

This thesis investigates tipping points in Software Engineering (SE) and exploreswhether concepts and methods from other complex systems domains can be appliedwithin a SE context. Software projects can experience sudden transitions from stabledevelopment into states characterized by instability, declining activity, coordinationproblems, or project abandonment. While tipping point theory and early warningsigns (EWS) have been extensively studied in domains such as ecology, climatescience, and finance, they have received limited attention within SE.A mixed methods approach combining a systematic literature review with empiricalanalysis of SE datasets was applied. The literature review identified tipping pointcharacteristics and tipping point types associated with SE activities. To evaluate theapplicability of existing tipping point detection methods, three open source softwaredatasets were analyzed using the ewstools framework.The results show that tipping point behavior in SE can be characterized using a setof EWS, where variance, autocorrelation, and skewness appeared most frequently.Their distribution closely aligned with findings from domains such as ecology andclimate science, suggesting that software projects may exhibit similar dynamicalbehaviors to other complex systems. Tipping point characteristics were identifiedacross multiple SE activities, particularly within Management and Requirementsrelated processes, highlighting the importance of socio-technical and organizationaldynamics in software project instability.The empirical analysis demonstrated that existing tipping point detection methodscan capture meaningful instability patterns within SE datasets. However, irregularand highly variable repository activity affected the consistency of several indicatorsacross project populations, suggesting that existing methods may require adaptationbefore they can be reliably applied within SE contexts.This thesis contributes a foundation for understanding tipping points in SE andprovides a basis for future research on instability detection and EWS tools withinsoftware projects.