Transfer Learning-based Emotion Detection System in Cultivating Workplace Harmony

The technological advancements made the organizations achieve more business profits and attain new horizons that significantly impact on employee stress and mental-well well-being. Moreover, due to the advent of Artificial intelligence (AI), which offers automotive tasks and efficiency gains, the employee get fear of losing their jobs. Due to this, the employee suffers from excessive stress and mental health issues. Therefore, emotion detection is imperative in such a stressful workplace environment. Inspired by this, we proposed a transfer learning-based emotion detection system to improve the workplace environment. For that, a facial emotion dataset is utilized, which comprises grayscale images of employee faces having emotions such as fear, neutral, anger, sadness, happiness, surprise, and disgust. Then, we used transfer learning-based pre-trained models with the intention of reducing the computational overhead of AI training. We employed ResNet, Inception, MobileNet, and EfficientNet which offer an effective accuracy while detecting the emotions of the employee. This strategic use of pre-existing models not only optimizes efficiency but also enhances the overall effectiveness of our facial emotion analysis system, ensuring a robust and accurate representation of diverse emotional states in employees. From the result analysis, we found that the ResNet outperforms other existing pre-trained models in terms of training accuracy (95.23%), training loss (0.41), and training time (2803 sec).