Assessing Basic Emotion via Machine Learning: Comparative Analysis of Number of Basic Emotions and Algorithms.

Abstract

This paper explores the use of machine learning (ML) methods to identify "clusters" of basic emotions based on pleasure, arousal, and dominance (PAD). The data was obtained from the Dataset for Emotion Analysis using Physiological Signals (DEAP), data collected within the Building and Designing Assistive Technology (BDAT) Lab using the International Affective Picture System (IAPS), and the scores of PAD from the IAPS. The objective is to develop an algorithm that maps a PAD score to clusters that express emotions, e.g., sadness or happiness. The elbow method was used to determine the optimal number of clusters (4-8), and nine different ML algorithms were compared. Decision Trees, polynomial support vector machines (SVMs) and linear SVMs provided accurate results. The Decision Tree demonstrated efficiency, during both testing and validation, in identifying the same clusters when analyzing both the DEAP and IAPS datasets. The dataset included limited data for each emotion creating the possibility of overfitting. However, when evaluating the results relative to previous research, the results added to the understanding of the nuances of emotion self-reporting and modelling.