Unsupervised Machine Learning Methods to Understand Relaxation Effect of Fragrances through EEG Technique

Objective: The objective was :1) To validate relaxation effects of fragrances proposed in the previous work using Unsupervised Machine Learning methods (USL). 2) To compare and quantify the relaxation potential of fragrances by establishing a criterion (metric). Methods: K-Means and Principal Component Analysis (PCA) were the employed USL methods. The data was a result of administering 4 essential oil fragrances (ALO (control), ECO, Lavender and AROMA) to 50 participants. PCA aided the characterisation of fragrances as function of 8 waves based on the relaxation induced. Results: The result from the previous work i.e., aromatherapy induced higher relaxation and, Alpha waves were good indicators of relaxation, was validated by K-Means. Higher Alpha wave intensity was associated with fragrance administration. Principal waves for olfactory stimulation were identified as Alpha, Beta and Theta. PCA analysis showed AROMA and Lavender fragrances had higher relaxation potential compared to the other fragrances. Weighted PCA showed the difference in the degree of relaxation for the administered fragrances. Conclusion: We concluded that Aromatherapy fragrance (a synergistic blend of relaxing essential oils) resulted in higher relaxed mental states. This was achieved by employing USL techniques as comparison and validation metrics for the previous study. Also, USL techniques were used to propose a methodology to understand and characterise EEG sensory data variability (subject-to-subject variability). The Relaxation Potential Metric successfully compared the degree of relaxation induced by olfactory stimuli. The significance of the proposed methodology is :1) It can be used as tool to analyse brain wave data from sensory stimulus/stimuli for better comparison and characterisation and 2) It can be employed to engineer products which are more consumer-centric in nature.