Innovative Emotion Recognition with CNN-SVM: Bridging Deep Learning and Statistical Classification

An attempt is made to classify the following five emotions using a CNN-SVM hybrid model designed for emotion recognition: the faces of fear, surprise, rage, happiness, and sorrow. Using convolutional layers, max-pooling layers, dropout layers, and fully-connected layers within the model’s architecture enables the extraction and classification of facial features via an efficient process. The model gives impressive results, as evidenced by the accuracy, precision, II, F value, and efficiency all over the spectrum, all exceeding for all classes. Mainly, Breaking class 3 shows the highest performance because it achieves exactness of $\mathbf{98 \%}$ concerning excellence at $\mathbf{92 \% 92 \%}$, recall of 97. The RNN network’s output reaches up to $\mathbf{58 \%}$, with precision levels the same as the F1 score of 95. 19%. This involves adding up macro, weighted and micro precision, recall and $\mathbf{F}$-Score metrics that equal $\mathbf{92 \%}$, which is the collective macro, weighted and micro averages-high AUC sleeves in the surroundings up to 1. Other incidents occurred at zero for all classes, leading to more evidence that the model is discriminatory. The model’s performance was not affected by differences between training and validation data as it scored $\mathbf{95 \%}$ and $\mathbf{93 \%}$ for the two categories, respectively, exhibiting high robustness and generalizability. By highlighting the superior performance of the CNN-SVM model in the emotion recognition domain, this study denotes a reliable remedy for real-life implementations.