Emotion Recognition Method based on Convolutional Neural Network and Black Widow Optimisation Algorithm

Abstract

Background: The automated Emotion Recognition (ER) method has gained popularity in several applications, namely, psychology and mental health. To accomplish this goal, deep learning algorithms are employed. These algorithms extract the appropriate features from the dataset images and recognise the emotions.
Methodology: In this paper, we have developed an emotion recognition method based on the Convolutional Neural Network (CNN). Furthermore, the main focus is on hyper-tuning the learning parameters of the CNN algorithm using the metaheuristic Black Widow Optimisation (BWO) to enhance the recognition accuracy. The BWO algorithm is based on the mating process of black widow spiders and provides a better convergence rate to find the optimal solution than other metaheuristic algorithms.
Results: The proposed method was simulated on the FER2013 dataset. The dataset was split into a 70:30 ratio. This reflects that 70% of the dataset was used for training purposes, whereas 30% of the dataset was utilised for testing purposes. The proposed method shows impressive results in recognising various emotions and achieves high values for the performance metrics, such as an average accuracy value of 0.99392, a precision value of 0.98467, a recall value of 0.98117, and an F1-score value of 0.96731. Finally, we have performed the comparative analysis of the presented approach with existing studies. The result shows that we have achieved better results due to employing the hyper-tuning strategy in the proposed method.
Conclusion and Recommendation: The proposed ER method is effectively recognising the different emotions. However, the same dataset was used for training and testing purposes, which negatively impacts the robustness and generalisation. Therefore, in the future, we will combine multiple datasets and balance them to evaluate the effectiveness of the proposed method.