Biserial targeted feature projection based radial kernel regressive deep belief neural learning for covid-19 prediction.

Coronavirus disease 2019 (COVID-19) is a highly infectious viral disease caused by the novel SARS-CoV-2 virus. Different prediction techniques have been developed to predict the coronavirus disease's existence in patients. However, the accurate prediction was not improved and time consumption was not minimized. In order to address these existing problems, a novel technique called Biserial Targeted Feature Projection-based Radial Kernel Regressive Deep Belief Neural Learning (BTFP-RKRDBNL) is introduced to perform accurate disease prediction with lesser time consumption. The BTFP-RKRDBNL techniques perform disease prediction with the help of different layers such as two visible layers namely input and layer and two hidden layers. Initially, the features and data are collected from the dataset and transmitted to the input layer. The Point Biserial Correlative Target feature projection is used to select relevant features and other irrelevant features are removed with minimizing the disease prediction time. Then the relevant features are sent to the hidden layer 2. Next, Radial Kernel Regression is applied to analyze the training features and testing disease features to identify the disease with higher accuracy and a lesser false positive rate. Experimental analysis is planned to measure the prediction accuracy, sensitivity, and specificity, and prediction time for different numbers of patients. The result illustrates that the method increases the prediction accuracy, sensitivity, and specificity by 10, 6, and 21% and reduces the prediction time by 10% as compared to state-of-the-art works.

Introduction

Prediction of coronavirus disease 2019 (COVID-19) is one of the major challenges in the world due to the rapid spread of the disease. Recent statistics designate that the number of people analyzed with COVID-19 is increasing exponentially and the disease is spreading to various countries across the world. The early prediction of theCOVID-19 is helping to minimize the mortality rate. There are various methods have been developed for Covid-19 prediction.

A Deep-LSTM ensemble model was developed in Shastri et al. (2021) to forecast the Covid-19 cases. The designed model increases the accuracy but the time consumption of disease prediction was not minimized. Cauchy Exploration Strategy Beetle Antennae Search and Adaptive Network-based Fuzzy Inference System (CESBAS-ANFIS) was developed in Zivkovic et al. (2021) to improve the prediction of Covid-19.CESBAS was used to solve other real-life NP-hard optimization problems. Swarm algorithms were applied to improve ANFIS time series forecasting. The proposed hybrid method was used to enhance ANFIS performance by determining its parameters via the CESBAS Meta heuristics approach.An enhanced beetle antennae search was employed to improve the overall performance of the prediction model. Though the designed method minimizes the mean square error, an efficient machine learning method was not applied for accurate classification, as well as for regression to minimize the prediction time.

Supervised machine learning techniques were introduced in Muhammad et al. (2021) for COVID-19infection. The performance assessment of the techniques showed that has better accuracy and sensitivity and specificity. However, the time consumption for Covid-19 prediction was not reduced. A Logistic model was introduced in Wang et al. (2020) for predicting the trend of COVID- 19 based on time series data. The designed model was not efficient for accurate prediction.

An artificial intelligence technique based on a deep convolutional neural network (CNN) was introduced in Alazab et al. (2020) to identify COVID-19 patients using real-world datasets. However, the designed technique failed to examine the effects of temperature on the COVID-19 patients. Machine learning techniques were introduced in Rustam et al. (2020) to predict the number of imminent patients influenced by COVID-19. The designed prediction methodology was not used as an updated dataset for accurate prediction through suitable machine learning methods.

Harris hawks optimizations (HHO) to optimize the Fuzzy K-nearest neighbor (FKNN) were introduced in Ye et al. (2019) to differentiate the risks of COVID-19. However, the designed algorithm archives higher prediction accuracy but the minimum error rate was not obtained. Boruta and Random Forest (RF) classifier were developed in Casiraghi et al. (2020) for fast and precise risk prediction of COVID-19 patients. However, the time consumption of the classifier was not minimized.

Three different selection models were developed in Walter Ageno et al. (2021) for forecasting the risk score to recognize patients at small risk. However, the designed models did not provide the better sensitivity to recognize patients at low risk. Artificial intelligence-mediated models were introduced in Suneeta Satpathy et al. (2021) to predict the mortality rate of COVID-19. The designed methods minimize the root mean square error. But the accuracy of prediction was not improved.

Most of the existing prediction methods have been designed for COVID-19. But, the accurate prediction was not enhanced and time consumption was not reduced. In addition, the updated dataset was less focused. Then, the existing prediction method failed to offer better sensitivity and specificity to recognize patients at low risk. But, the feature selection was not performed. To overcome the existing issue, Biserial Targeted Feature Projection-based Radial Kernel Regressive Deep Belief Neural Learning (BTFP-RKRDBNL) is introduced to enhance the accurate disease prediction with lesser time consumption.

The objective of the research work is as follows,

To obtain accurately predict the COVID-19 disease at an earlier stage, a novel BTFP-RKRDBNL technique is introduced.

To reduce the prediction time, the Point Biserial Correlative Target feature projection technique is used in the BTFP-RKRDBNL technique.

To enhance the accuracy of COVID-19 disease prediction, Radial Kernel Regressive Deep Belief Neural Learning is employed in the BTFP-RKRDBNL technique.

Major contributions

The major contribution of the proposed BTFP-RKRDBNL technique is explained as given below,

To improve the COVID-19 disease prediction accuracy, a novel BTFP-RKRDBNL technique is introduced using numerous layers with two different processes such as feature selection, and classification.

Radial Kernel Regressive Deep Belief Neural Learning is applied in the BTFP-RKRDBNL technique to estimate the weighted analysis between the testing and training value of the symptoms in the hidden layer of deep learning. This process improves the accuracy of COVID-19 disease prediction.

To minimize the prediction time, the Point Biserial Correlative Target feature projection technique is applied in the BTFP-RKRDBNL technique. The correlation between the features is estimated and obtains the results as dichotomous i.e. two output such as relevant and irrelevant. The relevant features are selected for classification and remove the other features.

Finally, an extensive experimental assessment of the BTFP-RKRDBNL technique is performed with two baseline prediction approaches for comparison. The BTFP-RKRDBNL technique is analyzed based on the various performance metrics.

Structure of the paper

The rest of the article is organized into different sections. Section 2 provides a related work of the prediction. Section 3 describes the explanation of the BTFP-RKRDBNL technique with a neat architecture diagram. In Sect. 4, an experimental assessment of the proposed and existing methods is performed with Novel Corona Virus 2019 Dataset. Section 4 provides a comparative analysis of the proposed and existing methods. At last, Sect. 5 provides the conclusion of the paper.

Related works

Three hybrid methods were introduced in Abbasimehr and Paki et al. (2021) for forecasting COVID-19 depends on deep learning models. However, the designed approach failed to obtain superior accuracy by extracting the useful features integrating them into the deep learning models.

A machine learning approach was introduced in Zoabi et al. (2021) for forecasting the COVID-19 based on signs. But, the designed approach failed to improve the performance of time consumption of prediction time. An Adaptive Neuro-Fuzzy Inference System (ANFIS) was developed in Celestine Iwendi et al. (2021) for early recognition of COVID-19 with higher accuracy. But the better sensitivity and specificity were not obtained.

Three different machine-learning models were developed in Assaf et al. (2020) to predict the risk level of Covid-19. However, the designed models provide a better performance of sensitivity, specificity, and accuracy but the time consumption analyses were not carried out. A hybrid artificial-intelligence (AI) method was introduced in Nanning Zheng et al. (2020) forCOVID-19 prediction. However, the higher accuracy of COVID-19 prediction was not improved. A hybrid model was designed in Luis Fernando Castillo Ossa et al. (2021) for forecasting the COVID-19. But the sensitivity and specificity analysis was not performed.

A novel parametric Suspicious-Infected-Death (SpID) model was developed in Tutsoy et al. (2020) for the prediction and investigation of the COVID-19 victims. The designed model failed to use the machine learning technique for accurate prediction with minimum time. The deep Learning method was developed in Connor Shorten et al. (2021) for COVID-19 applications. However, it failed to able to learn more samples with minimum time.

The artificial Intelligence method was introduced in Vaishya et al. (2020) to discover the diseases due to coronavirus and also used to monitor the condition of the patients. However, the performance of sensitivity and specificity remained unsolved. Deep learning with the statistical model was developed in Fokas et al. (2020) to yield better predictions of individuals infected with SARS-CoV-2. But the model failed to perform the significant feature selection.

Proposal methodology

A Coronavirus disease (COVID-19) is an infectious respiratory disease caused by the SARS-CoV-2 virus that is still spreading quickly in many countries and states worldwide. Therefore, it is urgent to conduct prediction research on the growth and spread of the epidemic. With the growth and spread of the epidemic, health care accurately performs the medical data analysis and early disease prediction. Moreover, the accuracy of a COVID-19 prediction is decreased due to a variety of symptoms. However, the existing machine learning works not efficient for accurate risk level prediction with minimal time. Therefore, a novel deep learning technique called BTFP-RKRDBNL is introduced for accurate Coronavirus disease prediction.

Figure 1 architecture and the process of the BTFP-RKRDBNL technique to obtain the accurate COVID 19 prediction using deep learning concept with higher accuracy and lesser time consumption. Let us consider the Novel Corona Virus 2019 Dataset for the prediction process. After that, the features and the data are collected from the dataset. Deep Belief Neural Learning includes two processes namely feature selection and classification. At first, the feature selection uses Point Biserial Correlative Target feature projection for selecting the relevant features and removes the other features. Then, the selected features are used for classification using Radial Kernel Regression. In this way, the accurate prediction is carried out with minimum time. The different processes are implemented in the deep neural network.

Fig. 1

Architecture of the proposed BTFP-RKRDBNL technique

Let us consider the Novel Corona Virus 2019 Dataset for the prediction process. After that, the features and the data are collected from the dataset. Deep Belief Neural Learning includes two processes namely feature selection and classification. At first, the feature selection uses Point Biserial Correlative Target feature projection for selecting the relevant features and removes the other features. Then, the selected features are used for classification using Radial Kernel Regression. In this way, the accurate prediction is carried out with minimum time.

Figure 2 illustrates the structural diagram of the Deep Belief Neural Learning for the classification of patient data. The structural diagram consists of numerous layers that comprise the neurons like the nodes. The nodes in one layer are connected to another layer and to structure the whole network. Deep Belief Neural Learning uses the two visible layers such as one input layer, one output layer, and two hidden layers. In the input layer, the number of features and the data are taken as input. The activity of the neuron in the input layer is expressed as follows,where ‘‘ denotes the activity of the neuron in input layer output. ‘‘ indicates the features, symbolizes the initial weight at the input layer, ‘‘ represents the bias stored the value is ‘1’. Then the input is transferred into the first hidden layer where the feature selection process is carried out.

Fig. 2

Structural diagram of Deep Belief Neural Learning

Point Biserial correlative target feature projection

The first process of the proposed BTFP-RKRDBNL technique performs the relevant feature selection in the first hidden layer using Point Biserial Correlation. The proposed BTFP-RKRDBNL technique uses a Point Biserial Correlative Target feature projection to perform feature selection. The Point Biserial correlation function is used to measure the correlation between the features. Then the higher correlated features are selected and other irrelevant features are removed. The relevant feature selection process of the proposed BTFP-RKRDBNL technique is to find the significant features from the dataset and remove the other features. This process minimizes the time complexity of classification. The point biserial coefficient is a correlation coefficient that helps to analyze the features and returns the output in terms of dichotomous i.e. two output such as relevant and irrelevant. Here the target is the relevant feature selection.

Let us consider the number of features in the dataset. The correlation is measured between the feature is estimated as follows,

From (1), denotes a correlation coefficient, denotes a probability of the feature to select as relevant or irrelevant, and denotes a deviation. Therefore, the deviation is measured as given below,where ‘‘indicates the total number of features, indicates a mean value, denotes features. The relevant features are identified based on higher correlation. In other words, the higher correlated features are selected for disease identification, and other features are removed. This process helps to minimize time consumption.

Radial Kernel Regression-based classification

After selecting the relevant features, the classification is done in the second hidden layer using Radial Kernel Regression. Regression is a machine learning technique to analyze the relationship between the testing and training data. Based on the analysis, the Covid 19 risk patients are correctly identified. The advantages of Radial basis function (RBF) networks are easy to design, good generalization. Regression is used to analyze the relationship between the testing and training data. The relationship is measured using Radial Kernel by applying the weighted analysis to obtain the final classification results. Finally, Covid-19 risk patients are correctly identified.

Let us consider the selected feature with patient data from the dataset. After collecting the data, the relationship is measured using Radial Kernel by applying the weighted analysis to obtain the final classification results. The advantages of Radial basis function (RBF) networks are easy to design, good generalization. Regression is used to analyze the relationship between the testing and training data.

The relationship is measured using Radial Kernel by applying the weighted analysis to obtain the final classification results. Finally, Covid-19 risk patients are correctly identified. The testing and training data analysis are obtained as follows,where indicates a Radial Kernel function, denotes a testing disease feature value, denotes a raining feature value, ‘‘ indicates a weight value. The Radial Kernel regression function provides the values ranges from 0 to 1. By setting the threshold, the patient health risks are analyzed. The output values that are lesser than the threshold are classified as the lesser risk patients. The output values are higher than the threshold are classified as heavy-risk patients. The output values are equal to the threshold are classified as the normal risk of patients. Let us consider the training and testing data of the patient such as age and symptoms namely body temperature, travel history, and chronic disease.

If the patient’s age is greater than 50 and the temperature is > 37 °C and the travel history is lesser than 14 days, and the chronic disease is present, then the patient’s health conditions are classified as heavy risk.

If the patient age is greater than 50 and the temperature lies between 37 and 39 °C and the travel history lies between 14 and 28 days and the chronic disease is present, then the patient health conditions are classified as ordinary or medium risk.

If the patient’s age is greater than 50 and the temperature is 37 °C and the travel history is greater than 28 days, and the chronic disease is absent, then the patient’s health conditions are classified as low risk.

In this way, all the patient data are correctly classified with higher accuracy and minimum error rate. The classified results are displayed at the output layer. The algorithmic process of the BTFP-RKRDBNL technique is described as given below,

Algorithm 1 describes the step by step process of BTFP-RKRDBNL technique to achieve higher prediction accuracy with minimum time consumption. The deep belief algorithm comprises numerous layers to learn the given input data. The features and data are taken as input to the input layer. Then the input is transferred into the first hidden layer. In that layer, the feature selection process is carried out in the first hidden layer.

The correlation between the features is measured using the Point Biserial correlation function. Then the higher correlated features are selected for classification. The selected features are sent to the second hidden layer. In that layer, the regression analysis is carried out to identify the patient’s risk level. Based on the regression analysis, the prediction level gets improved. This helps to improve the prediction accuracy and minimizes the error rate.

Experimental setup

Experimental evaluation of proposed BTFP-RKRDBNL technique and existing Deep-LSTM ensemble model Sourabh Shastri et al. (2021) CESBAS-ANFIS Zivkovic et al. (2021) are carried out in Java language using Novel Corona Virus 2019 Dataset is taken from the Kaggle [https://www.kaggle.com/sudalairajkumar/novel-corona-virus-2019-dataset]. The dataset has daily level information on the number of affected cases, deaths, and recovery from a coronavirus.

The dataset comprises the different CSV files. Among the files, we have taken the COVID19_open_line_list file for conducting the experiments. This file is downloaded from https://github.com/beoutbreakprepared/nCoV2019/tree/master/latest_data. The dataset consists of 33 features and 2,676,403 instances. The numbers of features are patient ID, age, sex, city, country, province, chronic_disease_binary, symptoms, travel history, and so on. Among these features, relevant features are selected for disease prediction.

Based on the objective of the proposed method (i.e., focused on accurate COVID-19 prediction with minimum time consumption) the existing methods such as the Deep-LSTM ensemble model CESBAS-ANFIS are taken as base paper. These two base papers are explained to understand the proposed method. The proposed method concept is derived by considering the problems of these base papers. The drawbacks of these methods are effectively convinced by implementing the proposed method.

Comparative analysis

The experimental results of the BTFP-RKRDBNL technique and existing Deep-LSTM ensemble model CESBAS-ANFIS are discussed. The parameters prediction accuracy, sensitivity, specificity, prediction time are used for analyzing the performance of the proposed method.

Based on the objective of the proposed BTFP-RKRDBNL technique, experimental parameters such as prediction accuracy, sensitivity, and specificity, and prediction time are selected for experimental purposes.

In our work, Point Biserial Correlative Target feature projection is used to select the relevant features and eradicates the irrelevant features. This process helps to reduce time consumption. Next, the Radial Kernel Regressive Deep Belief Neural Learning is employed to perform classification for accurately predicting the patient’s risk level. This is aids to enhance the prediction accuracy, specificity, and sensitivity.

Prediction accuracy

It is defined as the ratio of a number of data that are correctly predicted through the classification to the total number of patient data taken as input. The prediction accuracy is measured in terms of percentage (%).

The prediction accuracy is formulated as,where denotes a true positive, denotes a false positive, indicates the true negative, represents the false negative.

Sensitivity

The sensitivity is the ratio of true positive rate to the summation of a true positive and false negative during the classification. The formula for calculating the sensitivity is expressed as given below,where denotes a sensitivity, ‘‘ denotes a true positive, denotes a false negative. Sensitivity is measured in terms of percentage (%).

Specificity

It is defined as the ratio of actual true negatives and the summation of true negative and false positives.where denotes a specificity, ‘‘ denotes a true negative, represents the false positive. It is measured in terms of percentage (%).

Prediction time

It is measured as the amount of time taken by the algorithm for predicting the disease. Therefore, time is mathematically formulated as given below,where ‘‘ denotes the number of data. The prediction time is measured in terms of milliseconds (ms).

Table 1 reports the experimental results of disease prediction accuracy with respect to the number of patient data collected from the dataset. The prediction accuracy is measured based on the number of patient data taken in the ranges from 1000 to 10,000. For each method, ten different accuracy results are observed with various counts of input data. Among the three methods, the proposed BTFP-RKRDBNL technique provides superior performance than the other existing Deep-LSTM ensemble model CESBAS-ANFIS. When considering the 1000 patient data for calculating the prediction accuracy in the first iteration.

Table 1

Comparison of prediction accuracy versus the number of patient data

Number of patient data	Prediction accuracy (%)
	Deep-LSTM ensemble model	CESBAS-ANFIS	BTFP-RKRDBNL
1000	83	80	90
2000	85	81	90
3000	87.33	83.33	92
4000	85	82.5	92.5
5000	87	84	93.6
6000	85	82.5	93.16
7000	84.28	83.28	91.42
8000	85.87	83.75	91
9000	83.33	81.66	91.55
10,000	86	84	90.6

The proposed BTFP-RKRDBNL technique is correctly classified and the accuracy is 90% and the prediction accuracy of and is 83 and 80% respectively. Afterward, the various runs are carried out with the number of input patient data. Finally, the performance of the BTFP-RKRDBNL technique is compared to existing prediction methods. The performance results stated that the average results are taken into account, the proposed BTFP-RKRDBNL technique establishes a better performance than all other approaches.

The average results indicate that the prediction accuracy of the proposed technique is considerably increased by 8% when compared to other Deep-LSTM ensemble models and 11% when compared to CESBAS-ANFIS. The graphical representation of prediction accuracy is shown in Fig. 3.

Fig. 3

Visual representation of prediction accuracy

Figure 3 illustrates the graphical representation of the Covid-19 disease prediction accuracy versus a number of patient data. The prediction accuracy of three different prediction methods namely BTFP-RKRDBNL, Deep-LSTM ensemble model, and CESBAS-ANFIS is represented by three dissimilar colors like green, violet, and red respectively.

From the presented graphical results, it is seen that the BTFP-RKRDBNL technique slightly outperforms than the existing approaches. This is due to the application of the Radial Kernel Regressive Deep Belief Neural Learning. The kernel function analyzes the training data and the testing data. The displayed the patient risk prediction at the output layer with higher accuracy.

Table 2 provides the performance of prediction accuracy versus a number of patient data in the ranges from 1000 to 10,000. The observed results indicate that the BTFP-RKRDBNL technique outperforms well than the other existing methods. Let us consider the data samples as input, the sensitivity of the BTFP-RKRDBNL technique is 97.75%. Similarly, the sensitivity of the Deep-LSTM ensemble model and CESBAS-ANFIS are 95.06 and 92.30%respectively. Then the percentages of the sensitivity of the proposed technique are compare with existing results. The average values indicate that the sensitivity is significantly increased by 5 and 7% than the state-of-the-art methods (Table 3).

Table 2

Comparison of sensitivity versus the number of patient data

Number of patient data	Sensitivity (%)
	Deep-LSTM ensemble model	CESBAS-ANFIS	BTFP-RKRDBNL
1000	95.06	92.30	97.75
2000	94.11	91.87	98.29
3000	94.04	91.66	98.52
4000	94.11	93.16	98.63
5000	94.18	92.68	98.70
6000	93.12	90.90	98.36
7000	91.52	91.06	96.82
8000	92.72	91.04	97.62
9000	89.33	88.43	97.78
10,000	92.77	91.41	96.88

Table 3

Comparison of Specificity versus the number of patient data

Number of patient data	Specificity (%)
	Deep-LSTM ensemble model	CESBAS-ANFIS	BTFP-RKRDBNL
1000	68.42	63.63	72.72
2000	66.66	62.5	70.83
3000	47.91	50	68.96
4000	66.66	61.53	71.42
5000	57.14	55.55	70.27
6000	60.43	57.14	66.66
7000	54.54	55.08	57.14
8000	55.75	53.84	66.265
9000	46.66	48.48	66.666
10,000	47.05	48.64	64.70

Figure 4 demonstrates the comparison of the sensitivity of three prediction techniques BTFP-RKRDBNL, Deep-LSTM ensemble model, and CESBAS-ANFIS. For the different number of inputs, the various sensitivity results are obtained. The observed result confirms that the sensitivity of the proposed BTFP-RKRDBNL is better as compared to other techniques. The BTFP-RKRDBNL technique uses deep learning with a large number of data samples and correctly predicts the patient risk level at an earlier stage with the selected features. The deep learning technique correctly identifies the patient risk with higher accuracy and minimum error. This helps to increase the accurate disease prediction and improves the sensitivity.

Fig. 4

Visual representation of the sensitivity

The performance analysis of specificity versus a number of input data samples in the ranges from 1000 to 10,000. The specificity of the three methods is estimated based on the true negatives as well as false positives. Let us consider the 1000 patient data to measure the specificity for all three methods. By applying the BTFP-RKRDBNL technique, the specificity value is 72.72%. The specificity of the existing techniques Deep-LSTM ensemble model and CESBAS-ANFIS is observed by 68.42, 63.63% respectively.

The average of ten comparison results indicates that the specificity of the BTFP-RKRDBNL techniques significantly improved by 20% when compared to Sourabh Shastri et al. (2021), 22% when compared to Zivkovic et al. (2021).

Figure 5 illustrates the performance results of specificity along with the number of patient data taken from the datasets. The graphical visual results indicate that the BTFP-RKRDBNL technique increases the specificity when compared to the conventional prediction methods. By applying the deep learning classification, the testing and training samples are correctly analyzed with the help of the regression function. Based on the classification results, the patient risk levels are accurately predicted and minimize the incorrect prediction.

Fig. 5

Visual representation of specificity

The COVID 19 prediction time with the changing of the number of input data is given in Table 4.The tabulated results observed that the prediction time rises linearly with the increase of the number of patient data. Among the three methods, the BTFP-RKRDBNL technique minimizes prediction time than the other two methods. Let us consider the number of patient data is 1000.By using the prediction time formula in Eq. (7), the amount of time consumed to predict the disease is 0.028 ms to the total number of patients is 1000.

Table 4

Comparison of Prediction time versus number of patient data

Number of patient data	Prediction time (ms)
	Deep-LSTM ensemble model	CESBAS-ANFIS	BTFP-RKRDBNL
1000	30	34	28
2000	32	36	30
3000	36	39	33
4000	40	44	36
5000	43	45	40
6000	45	48	42
7000	47	49	44
8000	50	53	48
9000	54	57	52
10,000	58	60	55

The COVID 19 risk prediction time of the COVID 19 prediction time is and hence the risk prediction time of the other two conventional methods are and respectively with the similar input. For each method, ten results are observed with a different number of patient data.

The proposed method is able to provide better outcomes while changing experimental settings, i.e., the number of patients is considered as 1000–10,000.Therefore, the overall comparison results reveal that the patient risk prediction time is considerably reduced by 6 and 13% when compared to the Deep-LSTM ensemble model and CESBAS-ANFIS respectively.

Figure 6 shows the experimental results of the prediction time with respect to a number of data are taken from the dataset. The visual representation of the cone chart noticed that the BTFP-RKRDBNL technique outperforms well in terms of achieving lesser prediction time than the existing methods. This enhancement of the BTFP-RKRDBNL technique is achieved through the significant feature selection process.

Fig. 6

Visual representation of prediction time

The BTFP-RKRDBNL technique uses the Point Biserial Correlative Target feature projection. Based on the estimation, the positively correlated features are used to select significant features, and other irrelevant features are removed. With the selected significant features, the disease prediction is performed and hence it minimizes the prediction time.

Conclusion

A novel method called the BTFP-RKRDBNL technique is introduced for accurate COVID-19 disease prediction with minimum time consumption. The BTFP-RKRDBNL technique uses multiple layers for analyzing the given input patient data. Before the classification, the BTFP-RKRDBNL technique performs the significant feature selection using Point Biserial Correlative Target feature projection. The positively correlated features are used for the classification process hence it minimizes the prediction time. Followed by, the Radial Kernel Regression is applied to a BTFP-RKRDBNL technique for analyzing the testing and training data. The analysis results significantly show that the proposed technique accurately finds the risk level of the patient based on their symptoms. In this way, the accurate prediction is carried out with minimum time. LR-DDP method achieve accurate COVID-19 disease prediction with minimum time consumption and also increases the prediction accuracy,specificity, sensitivity and identify the patient’s risk level based on their symptoms. The experimental assessment of the BTFP-RKRDBNL technique and other existing methods is conducted with the novel COVID dataset. The numerical results and the performance discussion prove that the BTFP-RKRDBNL technique increases the prediction accuracy, sensitivity, specificity, and prediction time than the state-of-the-art methods. The proposed BTFP-RKRDBNL technique is designed for accurate COVID-19 disease prediction with minimum time consumption. In future work, the proposed BTFP-RKRDBNL technique is further implemented to accurate COVID-19 disease prediction with higher prediction accuracy and minimum time consumption by using artificial intelligence and soft computing for COVID-19 disease. The results prove that the BTFP-RKRDBNL technique attains better improvement of prediction accuracy sensitivity, and specificity by 10%, 6%, and 21% and minimize the prediction time by 10% as compared to another method.