Machine learning to predict person's interest towards visual object by utilizing EEG signal

This experiment aims to determine the best algorithm to predict person's interest towards visual object. EEG signal (signal captured by Electroencephalograph) is utilized to show if a person is interested. This can be useful, such as for marketers to discover consumer's brain response towards product and identify whether the product is interesting. Learning algorithm classifies subject's brain response into one of two classes: “interested” or “not interested”. Research on EEG signal regarding a person's interest is rare even no one explicitly discuss about it. It is vice versa for person's emotion or attention. That is why the observation of person's interest is done by adapting the handling on EEG signal that is usually used to identify emotion and attention. The handling includes EEG channels selection, learning algorithms selection, and frequency-based waves selection. This experiment, based on the handling, retrieves EEG signal from 12 EEG channels to get theta, alpha, and low beta wave. It involves the participants of 12 males and 20 females, 19– 22 years old. There are several algorithms that are fit and common to train EEG signal: SVM, ANN, KNN, and Naïve Bayes. To get the best algorithm, this experiment compares the performance measure of these algorithms, the configuration of parameter, and the characteristics. The result shows that SVM and ANN can be chosen as the best algorithms. SVM yielded accuracy of 83.65% and F-measure average of 83.70% by using kernel function of radial basis function. Meanwhile, ANN yielded accuracy of 83.36% and F-measure average of 83.30% by using 1 hidden layer, 50 hidden neurons, learning rate of 0.1, and momentum of 0.2. In addition to have optimum performance and slight difference, both algorithms also have several excellences.