Treatment of Imbalance Dataset for Human Emotion Classification

Abstract

Developments in biomedical science, signal processing technologies have led Electroencephalography (EEG) signals to be widely used in the diagnosis of brain disease and in the field of Brain-Computer Interface (BCI). The collected EEG signals are processed using Machine Learning-Random Forest and Naive Bayes- and Deep Learning-Recurrent Neural Network (RNN), Neural Network (NN) and Long Short Term Memory (LSTM)-Algorithms to obtain the recent mood of a person. The Algorithms mentioned above have been imposed on the data set in order to find out what the person is feeling at a particular moment. The following thesis is conducted to find out one of the following moods (happy, surprised, disgust, fear, anger and sadness) of a person at an instant, with an aim to obtain the result with least amount of time delay as the mood differs. It is pretty obvious that the accuracy of the output varies depending upon the algorithm used, time taken to process the data, so that it is easy for us to compare the reliability and dependency of a particular algorithm to another, prior to its practical implementation. The imbalance data sets that were used had an imbalanced class and thus, over fitting occurred. This problem was handled by generating Artificial Data sets with the use of SMOTE Oversampling Technique.