Electroencephalography based classification of emotions associated with road traffic noise using Gradient boosting algorithm

• Published in: Applied acoustics Vol. 206; p. 109306
• Main Authors: Manohare, Manish, Rajasekar, E., Parida, Manoranjan
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.04.2023
• Subjects: Electroencephalography (EEG); Emotion classification; Machine learning; Psychoacoustic indicators; Urban noise scenarios; Psychoacoustic indicators; Electroencephalography (EEG); Urban noise scenarios; Emotion classification; Machine learning
• ISSN: 0003-682X; 1872-910X
• DOI: 10.1016/j.apacoust.2023.109306

•A listening experiment is conducted with 14 participants who listened to 14 traffic noise stimuli.•EEG signals are recorded as a psychophysiological response to presented stimuli.•Noise indicators and Gradient boosting algorithmclassify five EEG-based emotion indexes.•High temporal variability makes acoustic environment stressful and vice versa.•Positively perceived temporal variation causes increase in interest. Traffic noise is one of the major contributors to the urban sound environment. Exposure to such sound causes a change in emotional and perceptual responses. Psychophysiological signals, subjective surveys, or laboratory-based experiments can be used to analyse these changes. This study uses a combination of psychophysiological (EEG) responses, noise indicators, and machine learning techniques to investigate the relation between road traffic noise indicators and change in perception. Noise signals representing different traffic scenarios were collected from New Delhi, India, from which 32 noise indicators were categorised as energetic, spectral and psychoacoustic indicators. Virtual reality listening experiments were administered with 14 subjects to capture the EEG responses. EEG responses for each audio-visual stimuli quantified five emotions: Engagement, Excitement, Interest, Relaxation, and Stress. The Gradient Boosting Model (GBM) and the Boruta feature selection algorithm are used to categorise noise indicators as explanatory variables and emotions as response variables. Engagement is classified most accurately (81.59%), followed by Excitement (63.28%), Interest (64.18%), Relaxation (71.53%), and Stress (72.24%). It is observed that loudness and spectral content have an impact on the engagement and excitement indices. Interest is impacted by temporal and spectral variation, whereas roughness level determines stress and relaxation state of the listener. In the Russell's circumplex model of affect, the perceptual response to traffic noise was located along the Stress and Relaxation dimension. When a noise signal exhibits significant temporal variation, the calming sound stimuli often turn out to be stressful. In order to transform chaotic scenarios to vibrant urban scenarios it is required to reduce the roughness and temporal variation in sound signal.