Affective-States-Classification

A neural network based on CNN-LSTM to predict the facial expression from MMA dataset which can classify 7 affective states such as angry, disgust, fear, happy, neutral, sad, and surprise. Features from the pictures automatically extracted in Convolutional Neural Network layer, ensued by a Long Short Term Memory neural network architecture to classify the facial expression from the input. This model works best because of CNN, it helps to fetch the most important features from the image. In MMA facial expression dataset, there are 3 directories - train, test, and validation, validation directory used for observing which model works better for a given dataset. Each of the directory has 7 sub-directories which consist of 7 distinguished facial expressions. In Train set, there are total 17366 images, whereas Test set encompasses approx. 92926 images, while validation set has 17356 instances.