Machine learning-based viewers’ preference prediction on social awareness advertisements using EEG

Introduction: One of the most promising applications of neuromarketing is to predict true consumer preference for advertisements to quantify their efficacy. Researchers have already established such neuromarketing systems for static advertisements and e-commerce products. However, more research is required to develop such a system for dynamic advertisements. In this study, we predicted consumer preference for awareness advertisements and explored neural clues that may generate new insights on how we can evaluate advertisements using neuromarketing techniques. Methods: We took 4 awareness topics and selected 2 advertisements from each topic, using 2 types of storytelling ('shock' and 'comic'), giving us a total of 8 advertisements. We prepared a custom 14-channel EEG dataset of 20 individuals watching these ads, along with their preferences and other self-reported measures. Machine learning was used to perform binary classification on viewers' preferences. Additionally, other markers, such as engagement index and alpha activity, were studied. Result: The highest average accuracy of 72% was achieved using the leave-one-ad-out method. Further analysis shows that the engagement index (beta/alpha + theta) or (beta/alpha) is an important indicator of self-reported ratings for these advertisements, which have been reported previously. Discussion: Our ML model outperforms the current state of the art in terms of model accuracy. Additionally, awareness advertisements were used for the first time for such a task since these advertisements are free from any sort of product or brand bias. This ensures that the preferences of the advertisements were solely on the design and storytelling.