Emotion in Pixels: Refining Expression Recognition with Hybrid Deep Learning

This paper describes the development of an enhanced face expression detection model made of CNN and Random Forest classifier. With a 1,860-picture dataset, our model was painstakingly trained to recognise and categorise five main facial expressions: This emotion set is made up of Neutral, Happiness, Anger, Fear, and Sadness. In the design of the layers, consecutive convolutional and pooling layers have been used with parameters that are obeyed by the input images to extract the most important features and subsequently, a Random Forest for classification as the final layer to obtain robust results. Thus, the general accuracy of the proposed model is as high as 89.24%, therefore, the usefulness of this model cannot be questioned. In addition to the model’s capacity to handle disparate facial expressions, the study explores the aspect of enhancement firstly in enhancing its underprediction of False Positives for complex emotions such as Fear. The main direction of the further work to increase accuracy and to improve the number of situations, where the proposed approach can be used, will be: widening the dataset, refining the architecture of the built model, and examining real-time processing.