Novel Deep Learning Framework for Facial Stress Detection Combined with Recommendation System

Stress has become an escalating problem in modern society and also poses a growing concern, jeopardizing welfare of the humanity. This common problem demands the integration of video cameras in numerous environments, detecting the stress via non-contact cameras as well as offering an assistance system arise a cost-efficient and prevalent accessible method for the users. The real-time videos hold a sufficient amount of raw data that can be extracted and transformed into pivotal information. This paper focuses on a novel workflow approach that comprises of combined convolutional neural networks with different architectures. Both CNN (alpha and ghost) are trained with distinct data from similar sources which have low to zero overlying in training samples. This paper is set to showcase an encouraging outcome utilizing various datasets by comparing suggested architectures and existing state-of-the-art (SoA) models. The purposed workflow generated an average increase of 5% in accuracy, detection, and recall compared to the SoA models.