A Machine Learning Approach for Artifact Removal from Brain Signal Sandhyalati Behera and Mihir Narayan Mohanty * Department of Electronics and Communication Engineering, ITER, Sikhsha ‘O’Anusandhan (Deemed to be University), Bhubaneswar, 751030, India *Corresponding Author: Mihir Narayan Mohanty. Email: mihirmohanty@soa.ac.in Received: 08 March 2022; Accepted: 27 May 2022 Abstract: Electroencephalography (EEG), helps to analyze the neuronal activity of a human brain in the form of electrical signals with high temporal resolution in the millisecond range. To extract clean clinical information from EEG signals, it is essential to remove unwanted artifacts that are due to different causes includ- ing at the time of acquisition. In this piece of work, the authors considered the EEG signal contaminated with Electrocardiogram (ECG) artifacts that occurs mostly in cardiac patients. The clean EEG is taken from the openly available Mendeley database whereas the ECG signal is collected from the Physionet data- base to create artifacts in the EEG signal and verify the proposed algorithm. Being the artifactual signal is non-linear and non-stationary the Random Vector Func- tional Link Network (RVFLN) model is used in this case. The Machine Learning approach has taken a leading role in every field of current research and RVFLN is one of them. For the proof of adaptive nature, the model is designed with EEG as a reference and artifactual EEG as input. The peaks of ECG signals are evaluated for artifact estimation as the amplitude is higher than the EEG signal. To vary the weight and reduce the error, an exponentially weighted Recursive Least Square (RLS) algorithm is used to design the adaptive filter with the novel RVFLN mod- el. The random vectors are considered in this model with a radial basis function to satisfy the required signal experimentation. It is found that the result is excellent in terms of Mean Square Error (MSE), Normalized Mean Square Error (NMSE), Relative Error (RE), Gain in Signal to Artifact Ratio (GSAR), Signal Noise Ratio (SNR), Information Quantity (IQ), and Improvement in Normalized Power Spec- trum (INPS). Also, the proposed method is compared with the earlier methods to show its efficacy. Keywords: Random vector functional link network (RVFLN); information quantity (IQ); constrained independent component analysis (cICA) 1 Introduction Electroencephalogram (EEG) is a non-invasive procedure of recording the neurophysiological activity of the brain signal by placing the electrodes over the scalp. Due to its non-invasiveness, the EEG signal is extensively used for the detection and prediction of neurological problems of the brain such as sleep This work is licensed under a Creative Commons Attribution 4.0 International License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. Computer Systems Science & Engineering DOI: 10.32604/csse.2023.029649 Article ech T Press Science