Intelligence Combiner: A Combination of Deep Learning and Handcrafted Features for an Adolescent Psychosis Prediction using EEG Signals
Ejay Nsugbe, Oluwarotimi Williams Samuel, Mojisola Grace Asogbon, Guanglin Li
Abstract
Schizophrenia is a psychiatric condition that affects a broad portion of the population and carries both health and economic consequences. EEG recordings, which involve the acquisition of brain waves non-invasively across the scalp of an individual, have been shown to be an effective means through which various neurodegenerative and psychiatric diseases can be diagnosed. The characterization and modelling of EEG signals involves the extraction of features, which in this area, have been mostly handcrafted based on expert knowledge. This work explores the comparison and performance of features extracted automatically via machine intelligence using convolutional neural networks (CNN), and a fusion of both handcrafted and CNN-based features towards yielding an enhanced feature set for enhanced prediction of schizophrenia in adolescents. The experimental results showed a greater ability to predict the adolescent psychosis for the fused set of features in the region of 98 % classification accuracy, and showcased a form of superior intelligence comprising of both handcrafted (expert intelligence) and CNN features (machine intelligence) for an enhanced prediction capability.