EmotiveNet: Five-Dimensional Emotion Detection with CNN-SVM Fusion

This paper encompasses meaningful discussions of affect recognition algorithms and a heterogeneous architecture that merges SVMs with CNNs. The model was mainly created using facial expression data to recognize five primary emotions: happiness or sadness, anger or fear, anger or fear, and amazement. Emotion recognition systems aim for a rise of accuracy and efficiency as the goal of their entity. CNN s are used in these models to obtain numerical information about emotional states from high-resolution face images. We employ the SVM (support vector machine) classifier to classify these characteristics into different emotional states accurately. Representing the 234×234 pixels as standards, a dataset of 2400 face pictures was utilized to compare the presented hybrid approach. Each image was thoroughly analyzed to ensure the method's viability. Precision, recall, F1-score, and total accuracy were used to compare the performance with the hybrid model's procedures. The hybrid model assayed the accuracy rate of 81.99%-95.33% and recall in the scope of 86.18%-97.3%, indicating excellent results. What was of great significance was that Class 3, related in most cases to the specific emotions, obtained the highest accuracy (97%) and F1-score (94.66%). Moreover, batteries that hold their performance for constant time were mentioned in the work that employed confusion and sophisticated matrices as well as weighted averages of micro, macro and cheap calculations. An average micro balanced in all parameters with a mean of 91.29%, implying that the detection range could be between various emotional states. This observation proves proof that the feedforward-backpropagation method's effectiveness could have numerous benefits in revolutionizing emotion identification science. The areas most likely to do so are interactivity, healthcare, and security.