EMOTIV-EPOC based electroencehalographic (EEG) responses to pleasant-unpleasant odors classification using machine learning algorithms

It is known that odor stimulus has effects on brain electrical activity. The odor effects on human emotions, behaviors and mood can not be denied. In contrast to other external stimulus, odor perception is a complex event due to its sensory and cognitive manners, so there is no exact statement for the effects of odors on human central nervous system. Some odor dysfunctions may appear as a result of some neurologic disease (Parkinson, Alzheimer, motor neuron disease, etc.). The sense of odor, ability to distinction different odors and recognizing odors can be decreased in these diseases. This situation is sometimes ignored in clinical treatment. There are bunch of electro-physiological methods to analyse brain electrical avtivity. One of these method is Electroencehalogram (EEG) that is known as a good source to comment about functioning of brain. In this study, it is aimed to analyze and classify the EEG responses related to pleasant – unpleasant odors. By the help of surveys belong to participant and graphs of power spectrum density, most dominant pleasant – unpleasant (two of each) odors were determined. Discrete Wavelet Transform (DWT) was applied to EEG odor responses and dimension of feature vectors was decreased by using some statistical operations. Multilayer perceptron, k-nearest neighbor, Naive Bayes and random forest algorithm were used as classifiers which belong to WEKA data mining program. Channel selection was performed to whole dataset by using differantial evolution algorithm. Classification procedure was repeated and the results were compared with previous ones. When using whole channels, NB gives 70.93 %, kNN gives 92.76 %, MLP gives 92.73 % and finally RF gives 99.19 % classification rate if we combine 2 best EEG subbands belong to each participant. In same manner, when using 5 selected channels, NB gives 68.45 %, kNN gives 88.95 %, MLP gives 88.83 % and finally RF gives 97.58 % classification rate. In present study, it is going to be examined which part of brain and frequency bands are responsible for odors. Besides, it is thought that proposed work is advisible to detect some neurological diseases in early stages.