Code Complexity Patterns for Mental Health Risk Prediction by Using Artificial Intelligence and Machine Learning

Software development is a cognitively intensive profession where tight deadlines, continuous integration cycles, and evolving requirements consistently generate stress, burnout, and psychological deterioration. Traditional mental health assessment tools — based on self-reporting and periodic clinical instruments are intrusive, retrospective, and fundamentally ill-suited to the continuous, large-scale monitoring requirements of modern software organisations. This paper proposes a non-intrusive, data-driven framework for predicting mental health risk in software developers by analysing code complexity patterns extracted passively from version control systems. The core hypothesis is that cognitive overload manifests measurably in coding behaviour: elevated cyclomatic complexity, irregular commit rhythms, increased code churn, and reduced modularity all correlate significantly with validated psychological stress indicators. A supervised machine learning pipeline is developed across structural, cognitive, and behavioural feature dimensions encompassing nine distinct metrics. The study recruited 120 software developers over a 12-week longitudinal period. Results show that a Deep Neural Network achieves 91.3% classification accuracy across three risk categories (Low, Moderate, High), with an AUC-ROC of 0.938.