EAX: An Emotion and Action Unit Extraction Framework for Real Time Depression Recognition

Depression, a ubiquitous disorder that affects many individuals, might cause unwanted consequences such as fatigue, stress, or suicidal ideation. To address this issue, predictive models have been developed to identify signs of depression early, enabling timely intervention and support. In this paper, we propose a framework to detect depression via emotion and Action Units extracted from a 2 -dimension facial contour representation from the FacePsy dataset. These features were subsequently fed into three machine learning models—Logistic Regression, K-Nearest Neighbors, and XGBoost—to differentiate between depressed and non-depressed individuals. Our experimental evaluation demonstrates that the proposed pipeline achieves an AUC score of up to $0.97 \pm 0.08$, underscoring its strong discriminative capability. In addition, our analysis shows the significant differences between the non-depressed group and depressed ones in emotion and facial activities. These promising results suggest that automated facial analysis can serve as an effective tool for early depression screening, paving the way for timely clinical intervention.