Prediction of Human Emotion using LeNet Architecture

This research explores the application of convolutional neural networks (CNNs) for emotion detection, focusing on the classification of “angry,” “happy,” and “sad” facial expressions. A Kaggle-sourced dataset was utilized to train the CNN model, which employed key hyperparameters such as a batch size of 64, an image size of $256 \times 256$, and a learning rate of 0.001. The proposed CNN architecture incorporated six filters with a $3 \times 3$ kernel size, dropout and regularization rates of 0.0, a stride of 1, and a $2 \times 2$ pooling layer. The fully connected dense layers consisted of 1024 and 128 units, respectively, with ReLU activation. The final output layer utilized a softmax activation function to classify images into the three emotion categories. The trained model achieved an impressive accuracy of $88 \%$, demonstrating the effectiveness of the chosen hyperparameters for emotion detection tasks. This research addresses the need for high-accuracy emotion recognition by introducing a modern CNN architecture and optimizing hyperparameters. The proposed model aims to surpass previous models by improving robustness across various datasets and classification accuracy.