Machine Learning Methods for Fear Classification Based on Physiological Features

Abstract

This paper focuses on the binary classification of the emotion of fear, based on the physiological data and subjective responses stored in the DEAP dataset. We performed a mapping between the discrete and dimensional emotional information considering the participants' ratings and extracted a substantial set of 40 types of features from the physiological data, which represented the input to various machine learning algorithms-Decision Trees, k-Nearest Neighbors, Support Vector Machine and artificial networks-accompanied by dimensionality reduction, feature selection and the tuning of the most relevant hyperparameters, boosting classification accuracy. The methodology we approached included tackling different situations, such as resolving the problem of having an imbalanced dataset through data augmentation, reducing overfitting, computing various metrics in order to obtain the most reliable classification scores and applying the Local Interpretable Model-Agnostic Explanations method for interpretation and for explaining predictions in a human-understandable manner. The results show that fear can be predicted very well (accuracies ranging from 91.7% using Gradient Boosting Trees to 93.5% using dimensionality reduction and Support Vector Machine) by extracting the most relevant features from the physiological data and by searching for the best parameters which maximize the machine learning algorithms' classification scores.

Keywords: emotion classification; emotion dimensions; fear classification; machine learning; neural networks.

Figure 1

Data processing flow.

Figure 2

DEAP trial session structure.

Figure 3

Fear classification process.

Figure 4

Dataset structure for non-overlapping approach.

Figure 5

Dataset structure for overlapping approach.

Figure 6

Learning curves, model scalability and model performance for the SVM algorithm, for the non-overlapping dataset.

Figure 7

Learning curves, scalability and model performance for the PCA + SVM algorithm, for the non-overlapping dataset.

Figure 8

Learning curves, model scalability and model performance for the Gradient Boosting Tree algorithm, for the non-overlapping dataset.

Figure 9

Learning curves, scalability and model performance for the RF algorithm, for the non-overlapping dataset.

Figure 10

Learning curves, scalability and model performance for the kNN algorithm, for the non-overlapping dataset.

Figure 11

Learning curves, scalability and model performance for the PCA + kNN algorithm, for the non-overlapping dataset.

Figure 12

Learning curves, model scalability and model performance for the SVM algorithm, for the overlapping dataset.

Figure 13

Learning curves, scalability and model performance for the PCA + SVM algorithm, for the overlapping dataset.

Figure 14

Learning curves, model scalability and model performance for the GBT algorithm, for the overlapping dataset.

Figure 15

Learning curves, scalability and model performance for the RF algorithm, for the overlapping dataset.

Figure 16

Learning curves, scalability and model performance for the kNN algorithm, for the overlapping dataset.

Figure 17

Learning curves, scalability and model performance for the PCA + kNN algorithm, for the overlapping dataset.

Figure 18

Accuracy and loss for Configuration 1, non-overlapping dataset.

Figure 19

Accuracy and loss for Configuration 2, non-overlapping dataset.

Figure 20

Accuracy and loss for Configuration 3, non-overlapping dataset.

Figure 21

Accuracy and loss for Configuration 4, non-overlapping dataset.

Figure 22

Accuracy and loss for Configuration 5, non-overlapping dataset.

Figure 23

Accuracy and loss for Configuration 1, overlapping dataset.

Figure 24

Accuracy and loss for Configuration 2, overlapping dataset.

Figure 25

Prediction features for situation 0 (no fear), for the non-overlapping dataset.

Figure 26

Prediction features for situation 1 (fear), for the non-overlapping dataset.

Figure 27

Prediction features for situation 0 (no fear), for the overlapping dataset.

Figure 28

Prediction features for situation 1 (fear), for the overlapping dataset.