A comparative study of machine learning and deep learning algorithms for recognizing facial emotions

This article compares the performance of three algorithms for Facial Emotion Recognition (FER). The algorithms chosen include one machine learning model - SupportVector Machine (SVM) and two deep learning models - Convolutional Neural Network (CNN) and VGG16. All three algorithms were implemented using Python and evaluated on FER2013 dataset. The metrics used for comparison were testing accuracy, training time and the size of the weights file. The best performing algorithm from the study is used to implement FER in real time on a live video stream. The implementation details are presented and results are discussed.

I. Introduction

Facial expressions are some of the best ways a person communicates with the outside world. It reflects the thoughts, beliefs and the emotions experienced by a person; thereby showcasing their mental state. FER can be defined as the process of detecting and recognizing a person's emotions with the help of their facial expressions [1] and hence can be employed for non-verbal communication [2]. With the recent advancements in humanoid robots, FER is utilized for multiple applications [3] [4]. These include lie detection, understanding certain mental disorders, analyzing customer reactions and many more.