Emotion classification using temporal and spectral features from IR-UWB-based respiration data

• Published in: Multimedia tools and applications 2023-05, Vol.82 (12), p.18565-18583
• Main Authors: Siddiqui, Hafeez Ur Rehman, Zafar, Kainat, Saleem, Adil Ali, Raza, Muhammad Amjad, Dudley, Sandra, Rustam, Furqan, Ashraf, Imran
• Format: Article
• Language: English
• Subjects: Classification; Columns (structural); Computer Communication Networks; Computer Science; Data Structures and Information Theory; Emotion recognition; Emotions; Galvanic skin response; Machine learning; Multimedia Information Systems; Principal components analysis; Respiration; Special Purpose and Application-Based Systems
• ISSN: 1380-7501; 1573-7721
• DOI: 10.1007/s11042-022-14091-5

Emotions play an important part in our daily lives since they influence our feelings and interactions. Because of the involvement of various response channels, emotion identification is a sophisticated and complex process. Emotion recognition has recently attracted the attention of the research industries as well as the scientific communities in the medical area. A variety of approaches have been explored to detect emotions from facial features, speech, text, and physiological signals. This study focuses on the classification of three basic emotions, i.e. happiness, disgust, and fear using a new set of spectral features extracted from the raw chest signal. We have used previously gathered data and biological signals (respiration rate). A structured dataset of 96810 records and 16 columns is maintained for experiments. The prior study’s accuracy was 76%, which served as the baseline. In this investigation, spectral and temporal features are used. Principal component analysis (PCA) is employed to select the best features. Extensive experiments are performed using several machine learning models for different scenarios of using all features, PCA-based selected features, and Chi-square selected features to analyze the efficacy of spectral features, as well as, feature selection approaches. Feature scaling is used to standardize the range of variables. Results indicate that using the PCA-based selective features, an accuracy of 98.52% can be obtained which is competitive with intrusive methods including electroencephalography and galvanic skin response signals while outperforming existing approaches.