A Comparative Analysis of Convolutional Neural Networks, Bi-Directional Gated Recurrent Units, and Bi-Directional Long Short-Term Memory for Sequential Data Processing Using Human Emotions Dataset

Emotion classification is an important field that can help healthcare professionals or researchers in their day-to-day activities when dealing with the emotional diagnoses of people. Not having an accurate classification might hinder the correct diagnosis of what a person is feeling, which also affects the work of healthcare professionals or researchers. Therefore, this study performed a comprehensive comparative analysis of three leading deep learning architectures: Convolutional Neural Networks (CNNs), Bidirectional Gated Recurrent Units (GRUs), and Long Short-Term Memory (LSTM) to know their effectiveness in sequential data processing using human emotion dataset guided by the CRoss Industry Standard Process for Data Mining (CRISP-DM) methodology. The models were evaluated using the accuracy, computational efficiency, and generalization ability matrix. Our results gave valuable insights into the optimal selection of neural network models for sequential data analysis, for emotion recognition. The GRU model had the highest accuracy of 92.90%, followed by the LSTM model with 92.70<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">%</sup>, and lastly, the CNN with 91.95%. This analysis aims to guide researchers and practitioners in the effective application of DL techniques to improve the accuracy and reliability of emotion-based systems based on sequential data.