Analysis of smoking addiction using machine learning methods with biological signals

Smoking addiction is considered a serious health problem worldwide. Smoking based on nicotine addiction affects millions of people and causes significant health problems. Scientific studies have shown that smoking affects brain activity. Electroencephalography (EEG) is a method used to measure brain activity. Brain waves are recorded through electrodes placed on the scalp. These recordings allow visualization and analysis of brain activity in different frequency ranges (such as delta, theta, alpha, beta, gamma). The main purpose of this thesis is to analyze smoking addiction with machine learning methods instead of traditional methods by extracting the time-frequency domain features of EEG data. In the first stage of the study, EEG data collected with visual stimuli from 30 different individuals were labeled with the Fagerström Nicotine Dependence Questionnaire (FTND). Later, EEG data was synthetically reproduced with GAN (Adversarial Generative Networks) to prevent the negativities of insufficient or unbalanced data in machine learning methods. The multiplied data were subjected to pre-processing such as Discrete Wavelet Transform (DWT), feature extraction in time and frequency domains. In the second stage, the pre-processed EEG data were classified with machine learning algorithms such as kNN (k-Nearest Neighbor), SVM (Support Vector Machines), ANN (Feed Forward Artificial Neural Networks) and RNN (Recurrent Neural Networks). The results showed that the brain responses to smoking stimuli were especially pronounced in the temporal and prefrontal lobe regions and higher classification success was achieved in the theta frequency band. The classification success rates made with feed forward ANN reached 99%. This study emphasizes the potential of using EEG-based machine learning methods in determining neurophysiological markers for smoking addiction.