Data Analytics on EEG Signal Features to Distinguish Familiar and Unfamiliar Information in Human Brain Memory

In forensic criminology, technology such as sensors has a crucial role in helping reveal a crime event. However, the sensor device for detecting honesty widely used today (polygraph) still needs to fully implement sensors to dig up information in memory and test testimony from people's statements. One of the sensors that can be used as an alternative to collecting information on human memory is the Electroencephalogram (EEG). Therefore, we propose a system that can test a person's statement regarding information familiar and unfamiliar to his views. This system utilizes EEG data to differentiate this information and several classifiers such as RNN, LSTM, and Bidirectional LSTM for the classification process. The EEG data used in this study were obtained through a recording process for 20 participants. The recording channels used are Temporal (T3, T4, T5, and T6) and Occipital (O1 and O2). In EEG pre-processing, several processes are needed, such as signal filtering, artefact removal, and band decomposition (Alpha, Beta, and Gamma). After obtaining clean EEG data from signal pre-processing, the signal feature extraction process is carried out in the time domain (skewness, kurtosis, and zero-crossing rate) and frequency domain (Welch's PSD). The signal feature values are input in several classifiers (RNN, LSTM, and Bidirectional LSTM). Based on the classification results, the best accuracy value was 87.94% for three conditions (baseline, familiar and unfamiliar) and 91.12% for two conditions (familiar and unfamiliar) using Bi-LSTM.