EEG mental arithmetic task levels classification using machine learning and deep learning algorithms

EEG signals are often-used modalities for monitoring mental arithmetic performance. In order to learn robust characteristics and predict the level of classification from EEG, we provide a deep neural networks and machine learning architecture. In this research, we considered online public dataset related to mental arithmetic task. EEG dataset recording are stored last 3 minutes of resting period (180 sec before task) and first 1 minute of the mental arithmetic task period (60 sec-while doing task). The task involves series of subtraction performed by 36 participants. The participants were divided among Good (1) and Bad (0) based on task level. We used machine learning and deep learning methods to classify three different task levels. (i.e., no task, normal task, and difficult task). In algorithm 1 we extracted five features like skewness, kurtosis, mean, standard deviation and variance from the EEG signals and applied as input to the classifiers. Here three machine learning classifiers used for three task levels classification like SVM, KNN and Decision Tree. KNN obtained high classification with 91% whereas SVM obtained S9% and Decision Tree obtained 65%. Algorithm 2 uses ANN and LSTM, two deep learning algorithms, to learn directly from raw EEG data. For better performance we used optimization techniques. ANN used rmsprop and obtained the classification accuracy of 96.80% whereas LSTM uses Adam and SGD and obtained the classification accuracy of 94%. Therefore, when raw data applied to deep learning algorithms can perform well when grouping of handmade features given to a machine learning algorithm.