Abstract:
Developer burnout has become a critical challenge in agile software development, as faster
delivery cycles, high task-switching demands, and sustained cognitive load affect mental wellbeing
and gradually reduce productivity. Although existing studies reveal that nearly 80% of
software developers tend to face burnout symptoms, burnout detection methods are often reactive
and solely rely on traditional surveys or isolated metrics. This study addresses this limitation
by proposing a proactive, all-in-one, datadriven framework that combines task management
data from Jira and code contribution activity from GitHub to predict burnout in agile teams. The
research problem is derived from the necessity for a unified system capable of detecting early
indicators of workload imbalance in an organization, answering stakeholder queries regarding
employee burnout, and supporting real-time intervention. The study extracts multi-modal metrics
from the Mozilla–Apache dataset, which consists of issue-tracking and version control data
collected from Jira and GitHub. Several types of machine learning models were trained and
evaluated to predict developer burnout. The Artificial Neural Network (ANN) revealed superior
predictive performance, obtaining 98.48% accuracy with a F1-score of 97.41%. Moreover,
correlation analysis indicated that factors such as commit count, total lines changed, issues handled,
and off-work time commits show the strongest associations with burnout risk.
