Transforming Facial Expression Prediction: Amplifying Accuracy with ResNet50 Features and Innovated XG-Boost Algorithm

• Main Author: M, Rajeshwari
• Publication Date: 2025
• Format: Article
• Language: eng
• Source: INFOCOMP: Jornal de Ciência da Computação
• Download full: https://infocomp.dcc.ufla.br/index.php/infocomp/article/view/3149

Facial expression is one of the most natural and a non-verbal way in expressing human emotions and interactions. Assessing the well-being via communication are the first signs that transmits the emotional state. This domain attracts more of research and are interested by the modality in the specificity of the domain. With a suspect able increase in the AI in domains, gaining a reasonable implementation of DL with its advanced ability in case of Facial Expression Detection known as FER. Existing studies adapting ML approaches, in the FER, have deployed in adequate laybacks such as high computational time, inability to deal with large datasets, and fail in bringing timely accurate ranges of predictions. In consideration to these aspects, the proposed study admits the system design, in aim of FER comprising feature extraction and the classification of the emotions using the DL models. These are done using the proposed approach uses Deep Multi-level feature extraction using ResNet50, which is more appropriate in optimal and exact feature selection mechanism. Followed by Weight-normalized XG-Boost classifier for the process of classifying various emotional expressions. This is adapted in aim, of maintaining the gradient descent step and admitting using larger dataset for learning. The input images are collected from the FER13, dataset consisting 28,709 sample image data and the test data consists of about 3,589 image data. These are initially pre-processed for better accuracy rates during feature extraction and classifications. The complete model in effective FER is evaluated using the performance metrics comprising Accuracy, Recall. F1-score and the precision rates. This analysis of the performance will aid in affirming the overall efficacy of the proposed system.