1. Introduction

COVID-19 is an infectious disease that was first time reported in China in December 2019 and spread rapidly around the world ^[1]. It affected the world economy and health enormously because of its transferable nature. The World Health Organization confirmed it was a pandemic. The symptoms of this pandemic are quite close to pneumonia ^[2]. If a patient is infected and has serious respiratory problems, then acute care is essentially required. To overcome the pandemic situation and to stop the further spread of the disease, researchers and medical experts are trying to find the appropriate vaccine. COVID-19 has become a challenge for all over the world. Some precautionary measures like partial or strict lockdowns halting community services like educational institutions, minimizing world economy, international and domestic travel, closing borders, etc. are adopted to overcome these challenges ^[3].

The economy of the world is at its adverse and the health sector is furiously dumped due to the prevalence of COVID-19. All countries consecrating precautions till the invention of medicine. Early detection of COVID-19 was necessary to overcome the disease ^[4]. Various studies have been conducted to diagnose coronavirus ^[5]. Different techniques based on artificial intelligence applications are performing their task to detect COVID-19 ^[6]. Medical practitioners as well as Artificial Intelligence (AI) experts, are working hard to control the vast spread of COVID-19 ^[7].

The researchers ^[8] explored an artificial intelligence-based methodology that is used for the identification of COVID-19 patients by utilizing chest X-ray (CXR) and CT scan images. The dataset used in the study contains images of Chest X-rays and infected images of COVID-19. A decision tree classifier was applied to detect coronavirus-infected persons. Different statistical parameters like precision, f1 score, and recall score were utilized to estimate the performance of the recommended method as well as the experimental results demonstrated high accuracy, robustness, and precision of the suggested technique. Similarly, the authors ^[9] highlighted an intelligent system to detect and diagnose COVID-19 that is empowered with deep learning techniques. The study consisted of a CNN approach for the identification of COVID-19 by using CT-scan and x-ray images. The experimental results showed 95.5% accuracy and study in contrast with other existing approaches. The researchers ^[10] stated that the use of the Computerized tomography (CT) method was found helpful in the early detection of the global pandemic COVID-19. The authors presented a novel fusing and ranking deep features to identify COVID-19. The dataset was classified with a support vector machine and the performance was more accurate.

The Deep learning (DL) technique consisting of a Convolutional neural network (CNN) can detect COVID-19 in terms of negative and positive. Deep learning delivers accurate results on CT images as compared to RT-PCR ^[11]. However, the CAD system faces the challenge of the shortage of a publicly accessible dataset of CT images. The researchers proposed a novel CAD system consisting of the fusion of multiple CNNs for the identification of COVID-19 ^[12]. In the First step, features were extracted individually from each network as well as a classifier named a Support vector machine (SVM) was utilized to classify them. In the next step, Principal component analysis (PCA) was used to deep features, and a support vector machine was trained through these feature sets. Then, chosen principal components were fused and contrasted with fused deep features obtained from each CNN. The experimental outcomes proved that the presented model was accurate and efficient in diagnosing COVID-19.

2. Related Work

Hall et al. ^[13] presented a technique consisting of a Deep convolutional neural network (DCNN) for COVID-19 detection. For the detection purpose, an approach DCNN was applied which used 10-fold cross-validation. The results to detect coronavirus give 89.2% precision.

Mohammed et al. ^[14] evaluated various Machine learning (ML) approaches like an artificial neural network, Decision tree (DT), SVM, K-Nearest neighbors (KNN), radial basis function (RBF), and different deep learning methods GoogleNet, Xception, ResNet50, MobileNets, and DarkNet for the recognition of COVID-19. A publicly available large X-ray image-based dataset was generated, and this dataset consisted of Normal and COVID-19-infected images. The COVID dataset consisted of a total number of 400 infected and 400 normal cases. The experimental outcomes illustrated that the ResNet50 model has achieved high accuracy as compared to seven conventional models based on ML and five pre-trained models consisting of DL.

Basu et al. ^[15] described an end-to-end two-stage framework comprising deep learning feature extractions and selection methods to detect COVID-19 using CT images. The proposed two-stage framework has achieved 98.87% accuracy for the 2926 CT scan dataset and 97.30% accuracy for the 2482 CT scan image dataset.

Kibriya et al. ^[16] described that Chest X-rays can play a significant role in the detection of COVID-19 at an early stage.

Mercaldo et al. ^[17] designed and implemented an automatic detection system based on DL for the identification of COVID-19. The suggested methods are based on transfer learning methods to identify COVID-19. In their study, they acquired a CT scan-based dataset from the Precision Medicine Department, at Campania University, Caser-ta, Italy. The dataset consisted of 400 CT scan images of 45 patients. The dataset was split into 80% and 20% whereas the proposed deep learning model was trained on 80% and tested on 20% of the dataset. A cross-validation approach with k=5 was adopted to assess the training process. The suggested method has achieved 95.0% accuracy, 95.0% recall, 95.0% specificity, and 95.0% F-measure.

Al-Waisy et al. ^[18] introduced a hybrid DL approach to identify COVID-19 disease using the COVID-CheXNet dataset that consisted of CXR images.

Mzoughi et al. ^[19] stated that most COVID-19 detection models often consist of deep learning approaches and attained a high success rate in the detection process. However, huge memory, overfitting issues, and more computational time are the main issues for using deep learning methods. To eliminate such problems, the authors proposed Efficient-Nets models integrated transfer learning to classify COVID-19 utilizing CXR images.

Sharma et al. ^[20] explored shallow architecture based on a Capsule network for the detection of COVID-19 through chest X-ray images. The investigational outcomes of the suggested model showed an accuracy of 97.69% for binary classification and an accuracy of 96.47% for multi-classification by using a 5-fold cross-validation technique.

Ullah et al. ^[21] exploited a dense attention mechanism consisting of DL for the recognition of COVID-19 in chest X-rays. Feature extraction by utilizing different techniques was adopted at various scales from input X-rays. The experimental outcomes proved that the suggested technique achieved state-of-the-art outcomes for COVID-19 detection with an accuracy of 97.22%, precision of 95.54%, specificity of 99.12%, and 96.87% sensitivity.

Narula et al. ^[22] presented a novel architecture consisting of CNN-integrated long short-term memory (LSTM) to detect COVID-19 utilizing a CT scan dataset. The suggested CNN-LSTM combinational model provided 98.91% accuracy, 100% specificity, 97.82% recall, 98.90 F1 score, and 100% precision.

3. The Proposed Scheme

The proposed Intelligent Prediction for Coronavirus using Computational Intelligence Approaches (IPC-FCIA) model is based on two sections namely training and validation sections shown in Fig. 1.

In the training section, the Internet of medical things (IoMT) is applied to obtain data from different sources in the form of X-ray images and features-based data. The dataset of ICU-admitted patients is acquired from Services Hospital, Pakistan. The features-based dataset consists of flu, diarrhea, fever, headaches, difficulty in breathing, tiredness, pneumonia, and chest pain, and on the other hand, the image-based dataset consists of affected COVID-19 X-rays and not affected COVID-19 X-rays. The proposed model used 80% (422) images for training and 20% (105) images for validation purposes from the image-based dataset. The proposed model used 80% (438) samples for training and 20% (109) samples for validation purposes from the features-based dataset. The data acquisition layer is used to attain both x-ray images and features-based data and is also stored in the database. After acquiring the data, the pre-processing layer activates, and different operations are performed on the data like noise removal and image resizing per-forms in x-ray images. Handling missing values and normalization technique performs in the features base data. To train the proposed IPC-FCIA model, a support vector machine is implemented on feature-based data, and a convolution neural network is implemented on an image-based dataset in the application layer. Subsequently, various statistical parameters like accuracy and miss rate were evaluated for the performance of the proposed IPC-FCIA model. The system needs to retrain the proposed model, when the learning criterion is not satisfied, and when learning criteria are satisfied then the decision-level function approach is applied to the trained model and trained data is also kept in the trained database of the cloud. Afterward the decision-level fusion, the model was again evaluated by using different statistical parameters. When the learning criteria are not satisfied then the system needs to be retrained by using an update rule for the inference engine and when the model satisfies the learning criteria then cloud cloud-infused database and validation section activates are used to store the trained data.

Fig. 1. Proposed IPC-FCIA system model.

Fig. 2. CNN model for prediction of COVID-19.

In the validation section, real-time data was obtained by using the IoMT devices for the evaluation of the proposed model. Captured data is stored in the real-time database and a fused trained model which is imported from the cloud for the prediction of coronavirus is used for evaluation of this data. When the proposed IPC-FCIA model, predicts coronavirus is negative then no need for further treatment, and if the proposed model predicts coronavirus is positive then referred for further treatment and data stored in the COVID-affected patient database in the cloud. Deep learning techniques provided significant results in the prediction of disease in every domain of life like agriculture, transport, aeronautics, etc. CNN is highly applied in image classification studies and is one of the variants of neural networks. It comprises of conventional layer, a pooling layer fully connected layers, and a classification layer. The recommended research has applied CNN for the detection of the coronavirus shown in Fig. 2. CNN consists of input, hidden, and output layers, and its layer structure is based on feature extraction and classification. Convolution, activation, and pooling layers are part of feature extraction and are fully connected, softmax is part of the classification.

The input layer takes images as input in the form of width, height, and depth. A $3 \times 3$ filter was applied to the image for the convolved purpose and after convolution, the ReLU activation function was applied to normalize the values. Subsequently, subsampling of the image is performed by using the pooling layer which reduces the input image size but retains the important information that leads to fast computation. In the pooling layer, the size of the input image is decreased by using the max-pooling layer. In the classification phase, the last pooled feature map was flattened in the form of a vector and fed into the last layer which is named as Fully connected (FC) layer. In the FC layer, every neuron is associated with another neuron of the next layer. The softmax function is applied to the output layer of the fully connected layer and converted all the predicted values in the vector form.

The proposed IPC-FCIA model is based on decision-level fusion empowered with fuzzy logic. Fuzzy logic deals with uncertain situations. The proposed IPC-FCIA for the decision-level fusion model is mathematically shown in Eq. (1).

Table 1 represents the mathematical and graphical membership functions of the fused computational intelligence approaches. Fig. 3 shows the lookup diagram for decision-level fusion to diagnose COVID-19.

${\pmb{R}}^{\pmb{\mathrm{1}}}_{\pmb{ipc}} =$ If ${\mu }_{CNN}\left(c\right)\ $is Negative and ${\mu }_{SVM}\left(s\right)$ is Negative, then the COVID-19 diagnosis is Negative.

${\pmb{R}}^{\pmb{\mathrm{2}}}_{\pmb{ipc}} =$ If ${\mu }_{CNN}\left(c\right)\ $ is Negative and ${\mu }_{SVM}\left(s\right)$ is Positive, then the COVID-19 diagnosis is Positive.

${\pmb{R}}^{\pmb{\mathrm{3}}}_{\pmb{ipc}} =$ If ${\mu }_{CNN}\left(c\right)\ $is Positive and ${\mu }_{SVM}\left(s\right)\ $is Negative, then the COVID-19 diagnosis is Positive.

${\pmb{R}}^{\pmb{\mathrm{4}}}_{\pmb{ipc}} =$ If ${\mu }_{CNN}\left(c\right)\ $is Positive and ${\mu }_{SVM}\left(s\right)$ is Positive, then the COVID-19 diagnosis is Positive.

The rules of the proposed IPC-FCIA model are represented as a fuzzy relation defined Eq. (2) below.

Defuzzifier is one of the most vital parts of the decision expert system. It is the method of mapping the fuzzy to the crisp output values. In the proposed IPC-FCIA system model to detect COVID-19 disease, the center of gravity defuzzifier is applied. The center of gravity defuzzifier identifies the ${\yen}*$ as the center of the area covered by the membership function of € that is shown in Eq. (3).

Fig. 4 illustrates the fused output of the detection for COVID-19 consisting of CNN and SVM classifier. If CNN has 0 to 0.5 and SVM has 0 to 0.5, then there is no detection of COVID-19, which represents a COVID-Negative. If CNN has 0.4 to 1 and SVM has 0.45 to 1, then COVID-19 is predicted as Positive, and for all other remaining conditions, COVID-19 detection is Positive.

Fig. 3. Lookup diagram for the proposed IPC-FCIA model.

Fig. 4. Rule surface for proposed IPC-FCIA model.

Table 1. Membership functions for used computational intelligence approaches.

4. Results and Discussion

The proposed IPC-FCIA model based on computational intelligence approaches used MATLAB 2020a for simulation and results a deep learning approach has been applied to the dataset which contains images for the detection of coronavirus, besides another approach SVM has been applied to the feature-based dataset for the detection of coronavirus. Fused computational intelligence approaches empowered with fuzzy logic have been applied to both approaches for the prediction of coronavirus disease. The proposed IPC-FCIA model used 80% (422) images for training and 20% (105) images for validation purposes from the image-based dataset. The proposed model used 80% (438) samples for training and 20% (109) samples for validation purposes from the features-based dataset. After decision-level fusion, a total number of 214 fused samples are taken from real-time data. These samples are utilized to assess the performance of the recommended IPC-FCIA model to predict COVID-19. Various statistical parameters are applied and compared with other existing techniques.

The proposed IPC-FCIA model predicts the early diagnosis of COVID-19 disease resulting in COVID-negative and COVID-positive.

The proposed intelligent prediction model for COVID-19 by using CNN (IPC-CNN) system for training purposes is shown in Table 2. A total number of images of 422 is required for training purposes which are further categorized into two classes COVID-positive and COVID-negative. For intelligent prediction in the COVID-positive category, 231 images were utilized. The proposed IPC-FCIA model correctly predicted 226 for COVID-19-positive and wrongly predicted 05 for COVID-negative. For prediction in the COVID-negative category, 191 images were utilized. The proposed IPC-CNN model correctly predicted 188 for COVID-negative and wrongly predicted 03 for COVID-positive.

The proposed intelligent prediction model for COVID-19 by using CNN (IPC-CNN) system for training purposes is shown in Table 2. A total number of images of 422 is acquired for training purposes which are further categorized into two classes named COVID-positive and COVID-negative. For intelligent prediction in the COVID-positive category, 231 images were utilized.

The proposed IPC-FCIA model correctly predicted 226 for COVID-positive and wrongly predicted 05 for COVID-negative. For prediction in the COVID-negative category, 191 images were utilized. The proposed IPC-CNN model correctly predicted 188 for COVID-negative and wrongly predicted 03 for COVID-positive.

The proposed intelligent prediction model for COVID-19 using a convolutional neural network (IPC-CNN) for validation purposes is shown in Table 3. The total 105 images acquired for validation were categorized into two classes named COVID-positive and COVID-negative. For intelligent forecast in the COVID-positive category, 57 images were utilized, for COVID-positive the proposed IPC-FCIA model correctly predicted 55 and for COVID-negative wrongly predicted 02. For prediction in the COVID-negative category, 48 images were utilized, for COVID-negative the proposed IPC-CNN model correctly predicted 47 and for COVID-positive wrongly predicted 01.

The proposed intelligent prediction for COVID-19 using the support vector machine (IPC-SVM) model for training purposes is shown in Table 4. For training purposes, a total number of 438 samples are obtained which are further categorized into two classes named COVID-positive and COVID-negative. For intelligent prediction in the COVID-positive category, 265 samples were used. For COVID positive, the proposed IPC-SVM model correctly predicted 262 and wrongly predicted 03 for COVID-negative. In the COVID-negative category, 173 samples were utilized for detection, the proposed IPC-SVM system correctly predicted 170 for COVID- negative and wrongly predicted 03 for COVID-positive.

The proposed intelligent prediction model for COVID-19 uses a support vector machine (IPC-SVM) for validation purposes shown in Table 5. A total number of samples 438 is acquired for validation. A total number of samples 109 are further categorized into two main classes which are named COVID-positive and COVID-negative. In the COVID-positive category, 66 samples were utilized for intelligent prediction, the proposed IPC-SVM model correctly predicted 64 for COVID-positive and wrongly predicted 02 for COVID-negative. For prediction in the COVID-negative category, 43 samples were utilized, the proposed IPC-SVM model correctly predicted 42 for COVID-negative and wrongly predicted 01 for COVID-positive.

The proposed IPC-FCIA model decision matrix for real-time data is shown in Table 6. Real-time data consisted of images and features-based data. A total number of real-time data 214 is obtained for validation purposes which are further categorized into two classes which are named COVID-positive and COVID-negative. For intelligent prediction in the COVID-positive category, 123 samples were utilized. For COVID-positive the proposed IPC-FCIA model correctly predicted 120 and for COVID-negative wrongly predicted 03. For prediction in the COVID-negative category, 91 samples were utilized, for COVID-negative the proposed IPC-FCIA model correctly predicted 89 and for COVID-positive wrongly predicted 02.

Table 7 represents the performance of various statistical measures such as accuracy, miss rate, true positive rate, true negative rate, false-positive rate, and false-negative rate for the proposed IPC-FCIA. The proposed IPC-FCIA model provides 97.66% and 2.34% accuracy and miss rate respectively for validation purposes. Various statistical measures like accuracy, miss rate, true positive rate, true negative rate, false-negative rate, and false-positive rate for the proposed IPC-FCIA model are 97.66%, 2.34%, 98.36%, 96.74%, 1.64%, and 3.26% respectively.

Fig. 5 represents the performance assessment of the suggested IPC-FCIA model in contrast with other existing techniques. This study is limited to convolutional neural networks, support vector machines, and fuzzy logic techniques.

Fig. 5. Proposed IPC-FCIA system model comparison with literature.

5. Conclusions

The rapid spread of the coronavirus has affected a large population of the world and has become a great threat to the lives of people. The use of the computerized tomography technique was more accurate and fast in providing better results to detect the coronavirus. Multi-input networks are used to detect coronavirus to enhance the performance of the proposed IPC-FCIA method. Firstly, CNN and SVM are applied to the dataset secondly, decision-level fusion by using the fuzzy logic technique applied to detect COVID-19. The proposed IPC-FCIA model for COVID-19 using fused computational intelligence approaches obtains an accuracy of 97.66% with a true positive rate of 98.36%, a true negative rate of 96.74%, a false-positive rate of 3.26%, and a false-negative rate of 1.64%. The proposed IPC-FCIA model achieves the highest benchmark accuracy when compared to other existing methodologies. The proposed model can be applied to other COVID-19 datasets for the evaluation of the performance of the IPC-FCIA model in future work.