Analyzing the Importance of Age-biased Data in Recognizing Emotions from Facial Expressions Using Custom Data Sets and CNN Algorithm

This paper explores the importance of age-biased data in recognizing emotions from facial expressions. A custom data set was created by separating existing data sets into adults and kids. Three CNN architectures were tested, and the SE-ResNeXt50(32×4d) achieved the highest accuracy at 79.42%. The age-based model outperformed the non-age-based model by 22.24%. The study highlights the impact of age-biased learning data and algorithm types on emotion recognition accuracy, particularly for fear and neutral emotions.