Evaluating the Impact of Gender on Facial Expression Recognition

Facial expression recognition (FER) is a critical technology with applications in healthcare, security, and media. However, variations in facial images and biased datasets, particularly concerning gender, often lead to inconsistent predictions. This research evaluates the impact of gender on FER using five deep learning models: Sequential, ResNet-50, DenseNet, MobileNetV2, and InceptionV3. These models were trained and tested on male-only, female-only, and mixed datasets to identify gender bias, determine which algorithms mitigate bias most effectively, and understand which facial expressions are more susceptible to bias. The study employs bias measures like accuracy difference, F-1 score difference and disparate impact. The findings aim to contribute to the development of fairer and more accurate FER systems by addressing gender bias and improving recognition across diverse demographic groups.