Skip to content
SDGs Center UDINUS

Evidence record · UDINUS · SDGs

Depression Classification Based on Facial Action Unit Intensity Features Using a CNN-Poolingless Framework

Rektorat UDINUS 2024 Published 2026-10-30

Evidence record · UDINUS · SDGs

Unit
Rektorat UDINUS
Period
2024
Goals
Record
bb6f087b-2e69-43d8-844a-d7ed168561d6
Published
2026-10-30

Universitas Dian Nuswantoro conducted research on depression classification based on facial Action Unit intensity features using a CNN-Poolingless framework. The activity involved UDINUS researchers Sugiyanto, I. Ketut Eddy Purnama, Eko Mulyanto Yuniarno, Wiwik Anggraeni, and Mauridhi Hery Purnomo as participants. The research addressed the need for early detection of depression because depression can interfere with work, education, and personal relationships and can increase the risk of suicide. The activity therefore examined a computational approach intended to support depression classification through facial expression analysis.

The research used 14 Action Unit intensity features in facial expressions. It considered ordinal coding of Action Units, which was described as more effective for facial expression recognition than binary coding. The proposed approach used an optimized lightweight convolutional neural network without a pooling layer. Grid Search was applied to obtain hyperparameter values for the number of filters, batch size, optimizer, and learning rate. The research examined the effect of omitting the pooling layer because pooling may remove important Action Unit intensity features and affect classification results.

Experiments were conducted using the DAIC WOz data set, which was prepared in 12 data variations. The framework was also tested using the CASME II data set for a model robustness test. The reported experimental results indicated that eliminating the pooling layer was worth considering across different experimental scenarios and data. The proposed framework performed better than frameworks that included a pooling layer and also showed better performance than other models in the comparison. The framework attained an accuracy score of 0.988, a loss score of 0.031, and an F1-Score of 0.991, indicating good precision and recall. This UDINUS activity contributes to the stated goal through research focused on depression classification and early detection using facial expression features.

Reference https://www.scopus.com/pages/publications/85201471680?origin=resultslist