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

Area of Science:

• Psychology
• Computer Science
• Affective Computing

Background:

• Emotion self-reporting and modeling are complex.
• Physiological signals and dimensional emotion models (pleasure, arousal, dominance - PAD) offer alternative approaches.
• Machine learning (ML) can analyze complex datasets for emotion identification.

Purpose of the Study:

• To develop and evaluate ML algorithms for clustering emotions based on PAD scores.
• To map PAD scores to discrete emotional states (e.g., happiness, sadness).
• To assess algorithm performance using established emotion datasets.

Main Methods:

• Utilized datasets from the Dataset for Emotion Analysis using Physiological Signals (DEAP) and the International Affective Picture System (IAPS).
• Applied the elbow method to determine optimal cluster numbers (4-8).
• Compared nine ML algorithms, focusing on Decision Trees and Support Vector Machines (SVMs).

Main Results:

• Decision Trees, polynomial SVMs, and linear SVMs showed accurate results.
• Decision Trees demonstrated consistent efficiency in identifying emotion clusters across DEAP and IAPS datasets.
• The study identified potential overfitting due to limited data per emotion.

Conclusions:

• ML, particularly Decision Trees, can effectively model and cluster emotions based on PAD dimensions.
• Findings contribute to a nuanced understanding of emotion self-reporting and computational emotion modeling.
• This approach offers a data-driven method for emotion classification.