Depression Severity Facial Action Units (AU) Feature Dataset

This dataset contains Facial Action Unit (AU) intensity features extracted from clinical video recordings of individuals diagnosed with varying levels of depression (mild, moderate, and severe), as well as healthy controls. The recordings were collected at Bahawal Victoria Hospital, Bahawalpur, Pakistan, under the supervision of certified psychiatrists. A total of 223,647 facial frames were processed using the OpenFace toolkit. From each frame, 35 AU intensity features (e.g., AU01_r, AU12_r) were extracted. These features serve as interpretable indicators for automatic depression severity classification using machine learning models. This public release includes:- A CSV file with AU features and corresponding class labels- Documentation describing preprocessing, labeling, and feature selection- A Python script to visualize AU distributions The dataset enables researchers to explore explainable AI models for depression diagnosis and is intended for use in academic and clinical research related to psychiatry, affective computing, and mental health diagnostics. **Labels** - 0: Healthy - 1: Mild Depression - 2: Moderate Depression - 3: Severe Depression **Total Samples:** 223,647 **Features:** 35 (AU Intensity Features) **Format:** CSV If you use this dataset, please cite it using the citation provided below and acknowledge the data collection site and contributors.