Objective: Schizophrenia is a form of psychosis which is prevalent within the population and affects both adults and adolescents, where it has also been noted that the conventional means towards diagnosing the disorder involve a high degree of subjectivity, can yield false positive results and are overall difficult to replicate. The use of electroencephalography (EEG) brain waves has been seen to carry promise as a means towards a non-invasive diagnosis of brain states via the acquisition of the bioelectrical manifestation of neuronal activities from across the skull. The acquisition of EEG signals involves a high number of electrode channels alongside the extraction of a relatively broad number of signal features which ultimately yields a high dimensional vector that scales upwards with the addition of data from subsequent subjects. Methods: As a means around dealing with this issue, this paper proposes a combination of signal decomposition prior to feature extraction, alongside dimensionality reduction, to obtain a low dimensional representation and embedding which is sufficient to allow for an improved prediction performance with a low dimensional projection capable of distinguishing between schizophrenic and non-schizophrenic adolescents. Results: The results from the pilot carried out, showed that a combination of a metaheuristic signal decomposition, linear dimensionality reduction and the long short-term memory deep learning model provided the best classification performance for the recognition of adolescent schizophrenia from a 1-dimensional embedding representation with a prediction accuracy of 88%. Conclusion: Further work in this area would now look into the application of probabilistic modelling as a means towards inferring an associated stage of the schizophrenia psychosis where predicted to be present in an adolescent subject.