Human Attention Recognition with Machine Learning from Brain-EEG Signals

Emotion recognition has always been a very popular field of research. Recently, EEG brain waves are used to recognize the emotional states of a person. Attention level also plays an important role in human life, but still demands more investigation. This paper proposes a noble human attention level recognition system using advanced machine learning algorithms. The system is cost-effective, single-channeled, and time-frequency scalp-EEG signals-based. In this study, the Bitalino EEG sensor board is used to record EEG signals from 30 human subjects in different attention states. Initially, the attention level was classified into three categories, which are focused, neutral and distracted. The data was taken while the subjects were watching interesting videos and boring lectures, doing simple and interesting math problems, and solving interesting and hard puzzles. At first, these EEG signals are pre-processed to remove noise such as muscle movement. Statistical coefficients (i.e. mean, standard deviation, skewness, kurtosis, and entropy) and statistical wavelet transform are used to extract meaningful features from the EEG signal. We mainly used two multi-scale wavelet packet statistics (WPS) and multi-scale wavelet packet energy statistics (WPES) to generate the feature vector. This feature vector was used to train the complex hybrid model with CNN and LSTM. This proposed method achieved almost 89% accuracy while determining the attention level of a subject.