Lie detection from EEG signals using deep learning algorithms

Deep learning algorithms are being used to classify electroencephalography (EEG) signals, resulting in the acquisition of various information about humans and the development of advancements that can improve human life. EEG is a method that monitors brain waves to understand the physiological and functional details and activities of the brain, encompassing neural activities occurring in the brain. This study involves the collection of EEG signals, their processing using signal processing techniques, and their classification using deep learning algorithms, with the aim of determining whether individuals' statements are true or false. The development of this study aims to achieve more reliable results compared to many existing methods used for lying detection. By classifying EEG signals using deep learning algorithms, it is possible to determine whether individuals' statements are true or false. During interrogations or in legal cases, EEG signals can be recorded while questioning guilty or innocent individuals. The detection of whether a person is lying or telling the truth can be accomplished by evaluating the EEG signals obtained from individuals using a trained model based on deep learning algorithms. In the scope of this thesis, EEG data has been collected, and a unique dataset called LieWaves, consisting of EEG lie data, has been created. Pre-processing techniques such as filtering, Independent Component Analysis (ICA), Artifact Subspace Reconstruction (ASR), and Automatic and Tunable Artifact Removal (ATAR) algorithms have been applied to remove artifacts from EEG signals. Feature extraction from EEG signals has been performed using Discrete Wavelet Transform (DWT) and Fast Fourier Transform (FFT) methods along with Statistical Methods (SM). Each feature vector obtained has been classified using Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), and CNNLSTM deep learning algorithms. In the analyses conducted using the ATAR + DWT + LSTM methods, the best performance was achieved with an accuracy rate of 99.88%. The methods used in this study are novel and applied for the first time on the LieWaves dataset. To the best of our knowledge, similar studies have not utilized ASR and ATAR methods in the pre-processing stage, and the CNNLSTM model has not been used in the classification stage. The results obtained demonstrate the success of these methods.