Abstract—Driver drowsiness is one of the major causes for several road accidents over the world. In this study, Electroencephalogram (EEG) signals were acquired using 14 electrodes from 50 subjects. All the electrodes are placed on the driver scalp based on International 10/20 standard and Butterworth 4 th order filter was used to remove the noise and artifact. Four EEG frequency bands (delta, theta, alpha, and beta) were analyzed on this work and extracted using Discrete Wavelet Packet Transform (DWPT). Fast Fourier Transform (FFT) was used to extract two statistical features such as spectral centroid and power spectral density (PSD) from the above frequency bands. Subtractive fuzzy classifier was used to map the extracted features into four different driver drowsiness levels namely, awake, drowsy, high drowsy and sleep stage1. As a result of this study points out the best average accuracy achieved by subtractive fuzzy inference classifier is 84.41% based on power spectral density feature extracted by “db4” wavelet function. Index Terms - EEG, Discrete Wavelet Transform, Fast Fourier Transform, Fuzzy inference system I. INTRODUCTION Electroencephalography (EEG) is one of the reliable and non invasive procedure for efficiently records the human brain activity. It has good temporal and spatial resolution and widely used in several applications which includes, drowsiness detection, attention detection, emotion and stress assessment, clinical diagnostics, etc [1-3]. Driver drowsiness and distraction are the most common reasons for several road accidents and becoming a critical issue in both developing and developed countries. In the literature, several research efforts have made on efficiently tracking the driver drowsiness through different modalities: physiological signals (Electroencephalogram, Electrooculogram (EoG)), vehicle behavior and driver physical response (head position, eye movements) [1] [2].In [2], the maximum mean drowsiness discrimination rate of 92% is achieved on discriminating three driver states namely; alert, Dr M.Murugappan is with the School of Mechatronics Engineering in Universiti Malaysia Perlis, Malaysia. (Phone: +60174064707; e-mail: murugappan@unimap.edu.my). Mousa K Wali is with the School of and Communication Engineering in Universiti Malaysia Perlis, Malaysia. (e-mail: musawali@yahoo.com). Prof Dr Badlishah R Ahmad is with the School of and Communication Engineering in Universiti Malaysia Perlis, Malaysia. (e-mail: badli@unimap.edu.my). Ms Subbulakshmi Murugappan is with the Institute of Engineering Mathematics, in Universiti Malaysia Perlis, Malaysia. (e-mail: subbulakshmi@unimap.edu.my). drowsy and sleep based on discrete wavelet transform (DWT) and EEG signals under four frequency bands (delta, theta, alpha, beta). Tsai et al. [3] designed a real time system which can discriminate alertness from drowsiness with accuracy of 79.1% and 90.91%, respectively based on back propagation neural network (BPNN). Torbjorn et al. [4] identify sleep stages 1, 2, 3 and 4 based on EoG signals. This present work has two main objectives: (1) to decide the best feature for efficient discrimination between different drowsiness levels, (2) to select the optimal wavelet function for getting the better classification accuracy. Audio-visual stimuli based virtual reality (VR) environment is created in our laboratory for mimicking the real world drive experience to the drivers. All the subjects are requested to drive the car monotonously for 1 hour to stimulate three different sleepy states such as drowsy, high drowsy, and sleep stage1 in addition to awake state. Two features namely; spectral centroid and power spectral density (PSD), are derived using wavelet transforms by applying four different wavelet functions (“db4”, “db8”, “sym8”, “coif5”) on four EEG frequency bands: (delta (δ)[0-4Hz], theta (θ)[4-8Hz], alpha (α)[8-12Hz], beta (β)[14-32Hz]) which required five levels of transform. Finally, subtractive fuzzy inference system is used to classify drowsiness levels. The rest of this paper is organized as follows: In Section II, we summarize the research methodology by elucidating the data acquisition process and Section III discusses about the preprocessing, and feature extraction using wavelet packet transform. Section IV illustrates the overview of fuzzy subtractive clustering and the results and discussion of this present work is given in Section V. Section VI concludes this work. II. DATA ACQUISITION The simulated driving environment used for this present work is shown in Figure 1. Infrared camera (IR) had been used to capture the driver face image for data validation after finishing the experiment. In this work, 50 subjects (43 Males and 7 Females) in the age range of 24 years to 34 years have participated. Emotiv EEG System is used to acquire the EEG signals over the complete scalp through 14 electrodes (FP1, FP2, F7, F8, F3, F4, T7, T8, P7, P8, O1, O2, A1, & A2). EEG signals are acquired at a sampling frequency of 128 Hz and band pass filtered between 0.05 Hz and 60 Hz. The reference electrode (A1) and ground electrode (A2) are placed on right and left ear lobes, respectively. The impedance of the electrodes is kept below 5 KΩ. Before start driving, the subject is asked to initially keep eyes closed for 2 min duration Subtractive Fuzzy Classifier based Driver Drowsiness Levels Classification using EEG M.Murugappan, Mousa K. Wali, R. Badlishah Ahmmad, and Subbulakshmi Murugappan 978-1-4673-4866-9/13/$31.00 ©2013 IEEE International conference on Communication and Signal Processing, April 3-5, 2013, India 159