Drowsiness Detection Based on EEG Signal Using Discrete Wavelet Transform (DWT) and K-Nearest Neighbors (K-NN) Methods

Abstract

Drowsiness generally occurs due to lack of sleep. Drowsiness can trigger various problems, such as decreasing productivity, damaging emotions, even to the point of causing serious accidents, both on the highway or in the workplace environment. One possible way to detect drowsiness is by using an Electroencephalographic (EEG) signal. EEG is a test used to evaluate the electrical activity in the brain. This research proposed a system that can detect drowsiness based on EEG signal using Discrete Wavelet Transform (DWT) as feature extraction and K-Nearest Neighbor (K-NN) as classification method of drowsy and normal conditions. At a preliminary stage, the system would perform a pre-processing to minimize noise signals using normalization and grounding magnitude. Feature extraction of these EEG signals was then decomposed using DWT function whereas the K-NN method is used to classify the EEG signals either in normal or drowsy conditions. The K-NN is done by Euclidean Distance Method. The private dataset consists of 60 signals, divided into 30 signals to normal and drowsy each. This research used DWT with eight-level decomposition of Alpha and Beta signals, and 3 wavelet family types (Coiflet, Symlet and Biorthogonal). Based on the results of tests conducted, EEG signals was decomposed using 3 different types of wavelet family generally provides accuracy values that are not much of a difference while selecting different K values for K-NN classification affects the accuracy. In conclusion, the value of k = 5 is the optimum value to classify normal dan drowsy condition. This condition is in accordance with the K-NN theory in which a greater k value can reduce noise in the classification process so it can improve accuracy of the system. This condition provides system performance with the highest accuracy around 90–100% for any type of wavelet family.