Detecting Facial Expressions

This paper shows that machine learning algorithms can be used in order to classify emotions into different categories. The most promising results were from the K-nearest neighbor model, with 61.5% accuracy in the training set, and 32.1% accuracy in the testing set.

The research question put forward by this paper is “To what extent can emotion detection artificial intelligence algorithms be used in order to improve education, business success, employee safety, healthcare and video game testing?” This paper looks to be able to give the consumers of the solution the ability to pinpoint which emotions educators, healthcare workers, video game testers and consumers of their business are feeling in order to improve their quality. In all of the areas that have been previously mentioned, there is a large inability to be able to identify when certain workers are feeling certain emotions, as a result of lack of recall memory, lack of honesty, or simply lack of awareness. In order to identify the satisfaction and dissatisfaction of healthcare workers, educators, video game testers and consumers of a business at certain moments, the quality of healthcare, education, video games and business websites will all increase respectively by being able to identify when workers are feeling a certain way, and how to reinforce or reduce the emotions being felt during those moments. In order to solve this problem, many classification algorithms will be explored, such as convolutional neural networks, K nearest neighbor, multi-layer perceptron classifier, amongst many others. These classification algorithms will look for a successful and quality classification of the images being presented in order to ensure excellence whenever this solution is implemented in the areas mentioned previously. By using the algorithms mentioned previously, the accuracy of the model was significantly higher than what it would have been had it been chosen randomly. For instance, with a convolutional neural network, there was an accuracy of 25%, whereas random assignment would have had a success rate of 14.29% The conclusions from this investigation are that artificial intelligence algorithms can be used in order to classify images into 7 categories according to the emotion being displayed on them, and this solution can and will be successfully implemented into the areas mentioned previously.

Explore More!

Research Paper