Classification of emotion-based body odour using machine learning approaches : A comparative study using differential mobility spectrometry (dms) data
Githaiga, Maureen (2024)
2024
Master's Programme in Computing Sciences
Informaatioteknologian ja viestinnän tiedekunta - Faculty of Information Technology and Communication Sciences
This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Hyväksymispäivämäärä
2024-11-18
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:tuni-2024111210122
https://urn.fi/URN:NBN:fi:tuni-2024111210122
Tiivistelmä
Emotions affect us in our daily lives, influencing how we communicate, the decisions we make,and our behaviour. The physiological responses triggered by these emotions, such as sweating, have become an interesting area of study, particularly in understanding how they manifest and what they reveal about different emotions. Differential Mobility Spectrometry (DMS), a form of electronic nose technology that mimics the human nose, has been applied to investigate the chemical composition of different Volatile Organic Compounds (VOCs). Machine learning algorithms have been leveraged successfully to analyze and distinguish different VOCs based on their DMS measurements.
This study aims to classify body odour from sweat samples using measurements from the DMS electronic nose. We seek to determine whether changes in sweat composition caused by exposure to emotion-evoking situations can provide insight into emotion-related responses.
Sixty composite super-donor samples were created from sweat samples collected from 16 participants while they watched neutral and fear-evoking videos. These samples were analyzed using DMS, which generates dispersion plots. The points of maximum intensity in the dispersion plots were used as input features. Various machine learning algorithms, including Support Vector Machines (SVM), Linear Discriminant Analysis (LDA), Decision Tree, Random Forest, and k-Nearest Neighbor Searching (k-NN), were employed. Leave-one-out (LOOCV), leave-one-participant out (LOPOCV), 3-fold and 5-fold cross-validation methods were used for evaluation.
The models showed high accuracy, with the SVM sigmoid kernel and LDA achieving 83% accuracy in LOOCV, k-NN achieving 82% in LOPOCV, random forest achieving 80% in LOOCV, 3-fold and 5-fold cross-validation, and SVM with RBF kernel achieving 79% in LOPOCV. LDA and SVM with the RBF kernel emerged as the top-performing models, demonstrating robust performance without overfitting, in contrast to others that exhibited varying degrees of overfitting. Overall, the models demonstrated a significant ability to distinguish between odours associated with fear and neutral emotional states.
The study’s findings suggest that machine learning algorithms can effectively identify patterns in sweat composition linked to different emotions, offering insights into how physiological responses relate to emotions. This study highlights the potential of electronic nose technology in emotion detection and contributes to the ongoing research on the relationship between emotions, physiological responses and body odour.
This study aims to classify body odour from sweat samples using measurements from the DMS electronic nose. We seek to determine whether changes in sweat composition caused by exposure to emotion-evoking situations can provide insight into emotion-related responses.
Sixty composite super-donor samples were created from sweat samples collected from 16 participants while they watched neutral and fear-evoking videos. These samples were analyzed using DMS, which generates dispersion plots. The points of maximum intensity in the dispersion plots were used as input features. Various machine learning algorithms, including Support Vector Machines (SVM), Linear Discriminant Analysis (LDA), Decision Tree, Random Forest, and k-Nearest Neighbor Searching (k-NN), were employed. Leave-one-out (LOOCV), leave-one-participant out (LOPOCV), 3-fold and 5-fold cross-validation methods were used for evaluation.
The models showed high accuracy, with the SVM sigmoid kernel and LDA achieving 83% accuracy in LOOCV, k-NN achieving 82% in LOPOCV, random forest achieving 80% in LOOCV, 3-fold and 5-fold cross-validation, and SVM with RBF kernel achieving 79% in LOPOCV. LDA and SVM with the RBF kernel emerged as the top-performing models, demonstrating robust performance without overfitting, in contrast to others that exhibited varying degrees of overfitting. Overall, the models demonstrated a significant ability to distinguish between odours associated with fear and neutral emotional states.
The study’s findings suggest that machine learning algorithms can effectively identify patterns in sweat composition linked to different emotions, offering insights into how physiological responses relate to emotions. This study highlights the potential of electronic nose technology in emotion detection and contributes to the ongoing research on the relationship between emotions, physiological responses and body odour.