Schizophrenia (SZ) is a harmful and prolonged brain disorder which affects normal speech, thinking, and the behavioral characteristics of a person. These days widely almost 1% of people affected mental disorder as per the report of World Health Organization (WHO) and the impact of this affect the people life. The symptoms of them are hallucination, disorganized discourse, and delusions [1]. Because of the problems in some brain functions [2], schizophrenia easily happens and people in different sectors have been caused by this disease [3,4]. To an individual, this disease cause damage in a micro-level and to the country this disease has affected a big cause in macro-level which affects the country economic system also. The person with symptoms of psychotic disorder shows impacts in their daily routine [5].

Therefore, the diagnosing of schizophrenic patients is becoming the serious topic. EEG is a low-cost and an effective device to record the activity of the brain [6,7]. Now-a-days, EEG has been largely exploited in the diagnosis of several nervous system diseases like Alzheimer’s disease, epilepsy, and schizophrenia. EEG can be implemented by placing the electrodes on Schizophrenia patient’s scalp at different location [8]. Based on the analysis, the existing technique is required to provide effective outcomes in terms of effective diagnosing and cognitive analysis with low resolution.

EEG provides detailed information about the disease by analyzing the characteristics of the signal transformation [9]. Therefore, EEG is considered as an efficient tool for predicting neurological disorders [10]. The EEG measurement setup shown in Fig. 1 is based on 64 channels. Several schizophrenia based analysis are processed which includes classification and interpretation based on neurological disorder [11]. This can be done through the utilization of machine learning (ML) and Deep Learning (DL) techniques.

ML algorithms have been used for SZ classification which is primarily based on traditional classifiers such as Decision Tree (DT), Linear Discriminant analysis (LD), K-Nearest Neighbor (KNN) and kernel discriminant analysis (KDA) [12,13]. These methods require large datasets for processing. In the applications of neuroscience, the advancement of DL shows the promising results, therefore, the DCNN is used as proposed methodology for this research work.

The major contributions of the works are:

19-channel EEG signals are utilized for both schizophrenia class and normal pattern.

This section specifies the existing problems in EEG based signal conduction system for predicting the Schizophrenia detection.

All the brain monitoring functions may fail to provide detailed information about the diseased portion; therefore, there is a requirement to utilize the EEG modality which is highly recommended. Moreover, cost is also affordable for diagnosing disease. So there is a necessity to enable an effective technique for enhancing the performance of the EEG based diagnosing modality with high spatio-temporal resolution data.

This section focuses on literature review related to signal conducting with EEG data for Schizophrenia disease prediction. For this analysis, the articles from 2016 to 2021 are collected and analyzed in terms of techniques used.

In 2020, Prabhakar et al. proposed four feature extraction methods such as Isometric Mapping, non-linear regression technique, partial least squares and expectation maximization based principal component analysis for classifying the SZ from EEG signals. Then, optimization algorithms were used for this feature optimization. Then the classification was done with the help of the Adaboost classifier and Naïve Bayesian Classifier. The implemented method showed its efficiency with a classification accuracy of 98.77%.

In 2019, Oh et al. [14] have presented a technique for SZ detection using a convolutional neural system. Here, the model collected the EEG signals from 28 people, where they were equally for healthy as well as SZ patients and created CNN model with 11 layers for verifying the signals. During the convolution stage, the features were extracted and the fully connected layer was used for classification. This model provides 98.07% of accuracy for healthy patient and 81.26% of accuracy for SZ patients.

In 2020, Sharma et al. [15] have developed schizophrenia detection by EEG signals with features of statistical and non-statistical. From a set of 16 electrodes, variance, higuchi fractal dimension, correlation and mean were calculated for these electrodes. Analysis was taken from each electrode to find a potential electrode set. From the features, SZ detection has been carried out by potential electrodes. Here the classification was done by Support Vector Machines (SVM) classifier and the results showed that SVM achieved a 100% of accuracy.

In 2019, Phang et al. [16] have proposed a DCNN for EEG classification to detect SZ. The features were hybrid by using CNN with multi-domain connection that contained one-dimensional and two-dimensional CNNs. An apparent group between healthy patient and SZ were identified by learning the representation of hierarchical latent from EEG signals [17]. When comparing with single-domain CNN, this model with combined connectivity achieved 93.06% of accuracy with the fusion of decision-level.

In 2016 Santos-Mayo et al. [18] used the wave of P3b and designed a Computer-aided diagnosis (CAD) system to identify the healthy and SZ. CAD system consists of EEG preprocessing [19,20], feature extraction, grouping of 7 electrodes, discriminant feature selection, and binary classification. Finally, the proposed method was compared over different ML classification methods [21] on the basis of certain measures.

In 2020 Sunil Kumar Prabhakar et al. [22] used four features such as lempel ziv complexity, detrend fluctuation analysis, largest lyapunov exponent and recurrence quantification analysis to classify the EEG signal for SZ. The results showed that a classification accuracy of 92.17% is obtained when Black-Hole (BH) optimization is utilized with SVM-Radial Basis Function (RBF) kernel.

Rumelhart et al. [23] defined Error back propagation (BP) as the popular algorithm for training DNNs. DNNs is a powerful technique for combining and generating information from multimodal data. CNN has demonstrated better performance in many computer vision tasks [24]. Hence, the CNN models are promising in processing clinical diagnosis and expression data to detect mental health conditions.

In 2021 Fu et al. [25] designed a novel Sch-net neural network based on a CNN, which perform the SZ speech detection using DL techniques. The proposed Sch-net combines the advantages of the two components, which can avoid the procedure of manual feature extraction and selection.

According to 10–20 international standards, horizontal and vertical eye movements of patients are recorded and extracted the EEG signals from 21 gold cup electrodes, where these were studied in Kim et al. [26].

In 2021 Zülfikar et al. [27] detect the SZ disease using the features of time frequency and Visual Geometry Group-16, a pre-trained model of CNN is used for training process and key features are extracted from scalogram images.

In 2015, Aharon et al. [28] presented the “TFFO” (Time-Frequency transformation followed by Feature-Optimization), a novel approach for SZ detection. The recordings for single node are suitable for this approach and process of data acquisition is feasible for the quick detection of SZ.

In 2020, Zulfikar et al. [29] revealed that there is a relationship between frequency components of an EEG recording and the SZ disease. In 2019, Torres Naira et al. [30] represented the relations between channels by using Pearson correlation coefficient and used the shorter matrix as an input to the CNN.

From the analysis of the several research articles, the existing techniques utilized for SZ disease are analyzed and understood for developing the novel EEG based diagnosing framework.

By implementing the combination of DCNN with Mayfly optimization, the proposed model produces better results. This section mainly focuses on providing the results obtained based on the implementation of the work in MATLAB platform. The following section provides results in terms of accuracy, loss, SNR, MSE and NRMSE etc. This shows that the proposed model is effective in obtaining the expected outcome.

The results obtained in terms of loss are also shown which depicts that the proposed model incurred minimum loss. The Tab. 2 shows the type and training process in terms of accuracy and loss.

The Fig. 4 shows the results in terms of input data provided with the mean of 193.2847 and StD is 115.0069.

The Fig. 5 shows the graphical image based on rank correlation with normalized data 0.9727.

In Fig. 5, the thick blue line represents smoothed and thin line represents unsmoothed during the training process. Similarly, in Fig. 6, the thick orange line represents smoothed and thin orange line represents unsmoothed.

The Fig. 6 illustrates the rank correlation ration based on training values and prediction values. The result obtained based on rank correlation is 0.9782.

The Fig. 7 provides information about the error evaluation based on input signals which obtains MSE = 2820.637, RMSE = 53.1097, NRMSE = 0.84011.

The Fig. 8 provides information about the error evaluation based on output signals which shows MSE = 823.9263, RMSE = 28.7041, NRMSE = 0.14949.

The Fig. 9 shows the Error Histogram Test which provides Error mean of 3.1232 and Error StD is 28.6597.

The Fig. 10 provides the graphical image of Error Histogram that provides error mean of 8.681, and error StD is 53.573.

The Fig. 11 shows that the graphical representation of function evaluation. The green graphical line indicates the estimated min objective and the blue line indicate that the min observed objective.

The Fig. 12 shows that histogram based on trained data. The results obtained in terms of error mean is 1.7184 and error StD is 17.8836.

The Fig. 13 shows that the rank correlation based on trained data. The result obtained for rank correlation is 0.70232.

The Fig. 14 shows graphical representation of regression output which shows that the proposed model provides better results. The small circle in diagram illustrates the number of data, the linear line indicates fit, and the dot line is relying just near the fit line.

The Fig. 15 shows the graphical representation of SNR with a result of 23.67 dB in which the blue line represents the fundamental, the orange line represents the noise, and the black line represents the DC and Harmonics.

The below Figs. 16 and 17 illustrates the graphical results of proposed technique based on accuracy and loss respectively. It shows the proposed model has obtained 95% of accuracy.

This section shows a comparative table obtained from different classifiers to prove the proposed model is better than other existing techniques. For this analysis, the classifiers based on detection of schizophrenia including SVM, SVM-RBF, LS-SVM, Decision Tree, KNN, DCNN, Proposed (CNN-MA) were used. The following Tab. 3 provides the comparative results based on proposed model.

As per the above mentioned comparative results table, the performance of the proposed model is highly effective in terms of performance evaluation metrics such as accuracy, precision, recall, and f-measure. The comparative performance results shows that proposed model is better than the existing techniques.

This proposed technique is based on signal conducting system with EEG signals for diagnosing the schizophrenia disease. The ultimate purpose of the proposed model is to develop a novel technique to provide automated prediction system that helps in classifying the affected and non-affected people by utilizing the hybrid DL technique. This technique is incorporated with an excellent optimization technique namely Mayfly optimization which incorporates the functionalities of PSO and genetic algorithm. Therefore, the proposed model produces effective results. The implementation of this model is carried out in MATLAB platform and the results are obtained. Moreover, the obtained results including accuracy, loss, MSE, RMSE, MRMSE, etc. show that the proposed model is very effective in providing better outcomes. By this analysis, the proposed model is proved to be a better technique in predicting schizophrenia disease.