Machine Learning Methods for Fear Classification Based on Physiological Features.

Area of Science:

• Affective computing
• Computational neuroscience
• Machine learning for emotion recognition

Background:

• The DEAP dataset contains physiological and subjective responses for emotion analysis.
• Accurate emotion classification is crucial for human-computer interaction and mental health applications.

Purpose of the Study:

• To develop a robust method for binary fear emotion classification using physiological signals.
• To explore the effectiveness of various machine learning algorithms and feature engineering techniques.

Main Methods:

• Extracted 40 features from physiological data and mapped them with subjective emotional ratings.
• Applied machine learning algorithms including Decision Trees, k-NN, SVM, and artificial networks.
• Utilized dimensionality reduction, feature selection, hyperparameter tuning, and data augmentation for imbalanced datasets.

Main Results:

• Achieved high classification accuracies for fear detection, ranging from 91.7% (Gradient Boosting Trees) to 93.5% (SVM with dimensionality reduction).
• Demonstrated the effectiveness of feature extraction and hyperparameter optimization in improving emotion classification.
• Employed Local Interpretable Model-Agnostic Explanations (LIME) for model interpretability.

Conclusions:

• Physiological data, when combined with advanced machine learning techniques, can reliably predict the emotion of fear.
• The study highlights the importance of feature engineering and model optimization for accurate emotion recognition.
• The findings support the potential of computational approaches for understanding and classifying human emotions.