http://annalsofrscb.ro 17070
A NOVEL PLASMA GENERATION OPTIMIZATION-BASED LSTM FRAMEWORK FOR COVID-19 PREDICTION
T.Nagalakshmi
Assistant Professor, Department of Computer Science and Engineering, Koneru Lakshmaiah Education Foundation, Vaddeswaram, AP, India
ABSTRACT
COVID-19 epidemic now disturbs the whole world which rapidly spread across the country. Millions of people have been diseased. Furthermore, to limit the spread of the virus, economies have been shut down. Several scientists are presently working in numerous fields to tackle this epidemic and its environments. With a view to avoid death, it is of much importance to identify the future cases, the spread rate of virus and future safety measures in the field of medicine. To avoid the death rate and to arrange medical safety equipment, an accurate forecasting model is needed. It is very critical to conduct forecasting research on the improvement and spread of the epidemic. Therefore, the proposed work targeted at introducing a new COVID-19 forecasting model by using a Plasma generation optimization (PGO) optimized Long Short-Term Memory (LSTM) network. LSTM is a recurrent neural network that used forecasting frameworks. To achieve a higher prediction accuracy in COVID-19, the hyperparameters of the LSTM network tuned by using PGO optimization. The proposed model is applied in the temporal data of coronavirus spread from world health organization (WHO). The accuracy of the proposed PGO-LSTM model was better than other forecasting models. The proposed PGO-LSTM based predicting network is talented for a higher dataset.
Keywords : LSTM, PGO,WHO,COVID-19
Introduction
The new type of virus has diseased millions of people worldwide lately discovered coronavirus (CoVs) . It was first observed from China in December 2019, and then spread speedily to all over the world [2].it can be categorized as α, β, γ, ψ - viruses. The α - and β -type (CoVs) is caused through bats. The γ - and ψ -CoVs have been widely spread by pigs. It leads to respiratory and neurological infections with variable severity, from asymptomatic to severe.
Sometimes they result in severe problems, particularly in respiratory systems. The ailing person shows symptoms only after a long time is the main problem regarding this disease. The people with less immunity suffered from lung problems. Based on severity, risk grade is defined by the world health organization, namely, grade I(high risk), grade II(moderate risk) and grade III(minimum risk). For those, who are suffering from minimum risk grade, self-quarantine, and social distancing is suggested. But for those who are suffering from grade I and II an appropriate medical precaution is suggested.
Currently, scientists and medical peoples are emerged to find a vaccine for CoV. Since there is no perfect vaccine discovered for killing CoV. Various medical researchers are examining the different extents of the epidemic and find the outcomes to support humanity.
There is a demand for predicting the spreading rate to tackle the ongoing pandemic. Due to the non-linear and complex nature of the spreading rate, deep learning-based algorithms can be used to solve the issues. In order to foresee the effect of the disease on the community, predicting models are used. This can monitor to control the widespread.
Today, there have been a lot of interest to use a deep learning algorithm for prediction.
Deep learning model constructed by using more number input, output and hidden layers[1-4]. in conventional ANN model, do not consider temporal effects. But, recurrent neural network (RNN) consider temporal representation for processing sequential data. long short-term memory is a type of RNN named LSTM. LSTM commonly applied for forecasting problems[5-10].
In this work, propose a hybrid PGO optimized LSTM network for COVID 19 prediction.
the PGO optimization integrated to LSTM to find optimal window size and number of LSTM units. In LSTM, the prediction process neglects useful data when the window size is too small.
The network will be over fitted when the window size is too large. Here, PGO used to optimize the hyper parameters of LSTM.
The structure of this work is as follows. Section 2 presented a related work. In section 2, an overview of similar works is being presented. In section 3, both the PGO and LSTM combined model explained implementation results described in section 4. The work ends with the conclusion in section 5.
Related Work
Research works include the anticipation of future cases and its analysis of the variables which are liable for the spread of CoV. Earlier COVID-19 predicting is an emerging problem hence time series prediction complexities have been analysed thoroughly in our previous works.
Kumar et al have proposed a predictive analysis for COVID-19 prediction. They have worked on the data set collected from various countries of India, UK and Germany by May 2020 and apply standard forecasting models of ARIMA and FBProphet. The proposed model worked well for prediction but accuracy levels affected by lockdown policies. It can be solved by using ensemble models.
Hassam Tahir et al proposed. ResNet-101 based model for COVID-19 prediction. The proposed algorithm includes two steps of prediction. In the first step, feature extraction is done by principal component analysis. Then, ResNet-101 neural architecture applied for prediction. The implementation results on the dataset which includes 5003 E-registration slips show that the proposed method achieves 92% accuracy compared to other algorithms.
Ahmad Sedaghat et al have proposed an SEIR-PAD model for data processing. The proposed model contains 7-set of ordinary differential equations for predicting clinical data.
MATLAB simulation results indicate the higher efficiency of the proposed model with better suitability for managing COVID-19 pandemic in GCC countries.
Sina Ardabili et al have introduced an optimized neural network architecture for predicting COVID-19 in a global data set. The efficiency of the proposed method compared to other conventional methods using the parameters of absolute percentage error (MAPE) and correlation coefficient (r) values. From the results observed that the proposed technique can successfully cope with the forecast task. He, P et al have proposed a correlation function method for COVID-19 risk prevention and control. The correlation function method is a decision-making model working based on applying cross-correlation.
Nana Ramadijanti et al have presented a new model called SEIR epidemic. The proposed model analyzed with the parameters of cure rate, death rate and communication rate. The SEIR model trained by a number of cases that occurred on daily basis. Then, maximum correlation values used for prediction. Implementation outcomes show that the correctness of the SEIR technique is high when compared to other regression models.
Rajani Kumari et al have proposed a multiple linear regression model for COVID-19 prediction. In order to improve accuracy, autoregression has been used. Results conclude that lockdown and social distancing are identified as the best solution to stop the spreading rate.
Ryan Yixiang Wang et al have proposed a Neural Network model and Random Forest combined classifier for spreading rate prediction. Combining genetic variants with the individual's phenotypes used to form a hybrid model. The classification accuracy of 94.3 is achieved for various data sets.
Ruizhi Han et al have proposed of Broad Learning System (BLS) to forecast the death rate of COVID-19 patients. The proposed BLS system applied in blood samples of patients.
Implementation results show that the proposed model shows a sensitivity of 89.8% while having a specificity of 96.1%. The parameter of prediction correctness and confusion matrix used to prove the efficiency with other prediction models. Yunxiang Liu et al have proposed a prediction model using machine learning. The multi-objective support vector machine used for accurate prediction in Xinjian. The overall accuracy of 92.6 % achieved compared to other classifiers.
Onder Tutsoy et al have proposed a parametric Suspicious-Infected-Death (SpID) model for calculating COVID-19 casualties. This model finds casualties about 300 days whereas the count of doubtful people will need almost 1000 days to be reduced in the existing situations.
Prajoy podder et al have addressed the difficulties of applying machine learning algorithms for COVID-19 spreading rate prediction. Two types of classifiers used for managing coronavirus disease. Firs one for predicting the number of patients. The second one for predicting hospital requirements. The proposed model coded in python programming and applied in various COIVD data sets. The random forest (RF) based classifier achieves an accuracy of 98.13% with a recall of 92%. Safa bahri et al have aimed to apply Recurrent Neural Network architecture for predicting COVID-19 recovering cases in the USA. Long Short-Term Memory (LSTM) based RNN used for training and testing. Implementation results show that LSTM achieves higher accuracy with a minimum mean square error rate of 3%.
Hua Ye et al have proposed a severity classifying model for COVID patients. The proposed model uses Fuzzy K-nearest neighbour for feature discrimination. in order to tune the fuzzy parameters, the Harris hawks optimization (HHO) used as a fitness function. The HHO effectively choose subset for Fuzzy tuning. Implementation result displays that the proposed HHO-FKNN can get improved classification performance and higher stability to distinguish severe COVID-19 from mild COVID-19.
Nanning Zheng et al have proposed an improved susceptible-infected (ISI) model to classify the infection rate of COVID patients. The LSTM network combined with a natural language processing (NLP) module to construct a hybrid model for prediction. it also considers feedback of patients to retrain the network. Experimental results show that the proposed hybrid AI model can meaningfully decrease the errors of the forecast results and attain the mean absolute percentage errors (MAPEs) of 0.52.
Proposed Optimized LSTM PGO
Plasma generation optimization (PGO) is population-based optimization motivated by the generation procedure of plasma. It is used to solve engineering problems by the movement of electrons and their varying energy levels. Usually, it starts with a random population of electrons to attain the greatest degree of ionization. due to the two-phase search nature of PGO optimization, it is used to tune LSTM parameters
In PGO, the movement of the electrons iteratively continued until reach the highest energy as the best solution. It is organized into here steps. : Produce preliminary population of electrons, Simulation of d-orbitals in an atom, Calculating step size of electrons movement Step 1 The initial position is updated by the following equation
Si’j=Smin+rand (Sj,max-Sj,min);i=1,2,…..n
where Si,j is the primary location of the jth design variable for the ith electron. Here, Sj,min and Sj, maxes are the smallest and extreme allowable values for the jth design variable, correspondingly; rand is a random number that varies from zero to one
Step 2 This simulation is calculated by pear-shaped curves concentrating the resulting equation:
αi,j=U (0.6+0.1*d,1.4-0.1*d);
d=iteration/Max iteration where αi,j is the coefficient
Step3 To find the step size of an electron, a random is produced between zero to one IF rand<EDR Excitation is occurred
Otherwise, Ionization started LSTM
LSTM is a type of RNN with the additional element of memorization block to store the past event. LSTM can process with past data of more than 1000 time steps.it includes three types of multiplicative gating units; input, output and a forget gate. The LSTM block diagram shown in figure
The input value of LSTM is
It=𝜎 (Wixt+Utht-1+bi) Ct=tanh (Wcxt+Ucht-1+bc) The new state of a memory cell is updated as
Ct=it*ct+ft*ct-1; The output of a cell is termed as
ht=Ot*tanh (ct)
LSTM can handle a large amount of computational problem with minimum cost.
In this work, propose a hybrid PGO optimized LSTM network for COVID 19 prediction.
The PGO optimization integrated to LSTM in order to find optimal window size and number of LSTM units. In LSTM, the prediction process neglects useful data when the window size is too small. The network will be over fitted when the window size is too large. Here, PGO used to optimize the hyper parameters of LSTM.
The flow chart of proposed optimization is shown in figure 1. The movement of electron to get a highest energy is mathematically modeled to adjust the parameters of LSTM model. The proper tuning of parameters leads to higher accuracy.
Result and Discussion
The proposed framework is implemented in MATLAB and applied in the temporal data of obtained from WHO for various countries . For performance analysis, two performance metrics are considered; root-mean-squared error (RMSE) and mean absolute error (MAE). Both MAE
and RMSE used to measure the prediction inaccuracy or root mean square of the deviation from actual results.
Table 1 Performance measure.
Method MSE RMSE ACCURACY
LSTM 196.3 10.2 89
LSTM-LOA 181 9.5 93.5
PGO-LSTM 163.33 8.2 97.5
NO
COVID DATA
Training Set Test Set
Population Initialization
PGO Optimized LSTM
Optimal Values
Termination Criteria met?
Random Electron selection
Simulation of d-orbitals in atom
Step size adjustment
PGO
YES
Figure 1. Proposed LSTM-PGO
Figure 2 Performance analysis
The comparison result of the proposed model is shown in Table 1 and Figure 2, where the proposed PGO-LSTM is showed that it was much improved than the existing LSTM and LSTM- LOA(Lion optimization). In the MSE value, the LSTM has its value as 196.3 and LSTM-LOA has its value as 181 5, while the predicted MSE for the proposed technique is 163.33 which enhances the performances by 9.3% than existing. This result concludes that PGO-LSTM predicting with optimization surpasses the traditional model with higher efficiency.
Conclusion
There are numerous disadvantages when using a standalone LSTM model. It suffers from an over fitting problem and an infrequent distribution of input data. Hence, the plasma generation method is utilized to optimize the parameters of predicting network and introducing PGO- LSTM- network. The PGO method search for optimal point over different movement by the Simulation of d-orbitals in an atom. The proposed model is desired over conventional LSTM with the view of forecast correctness and convergence level.
References
[1]Ahmad Sedaghat; Shahab Band; Amir Mosavi; Laszlo Nadai ,COVID-19 (Coronavirus Disease) Outbreak Prediction Using a Susceptible-Exposed-Symptomatic Infected-Recovered- Super Spreaders-Asymptomatic Infected-Deceased-Critical (SEIR-PADC) Dynamic Model, 2020 IEEE 3rd International Conference and Workshop in Óbuda on Electrical and Power Engineering (CANDO-EPE).
[2] B.Sakthivel, R.Jeyapandiprathap, M.Jeyamurugan, G.Narmadha "Iot Based Solar Power Monitoring And Prediction Using Cuckoo Optimized LSTM" International Journal of P2P Network Trends and Technology 11.2 (2021): 6-8.
[3] Furqan Rustam; Aijaz Ahmad Reshi; Arif Mehmood,COVID-19 Future Forecasting Using Supervised Machine Learning Models, IEEE Access ( Volume: 8),Page(s): 101489 – 101499 [4] Hassam Tahir; Annas Iftikhar; Mustehsan Mumraiz , Forecasting COVID-19 via
Registration Slips of Patients using ResNet-101 and Performance Analysis and Comparison of Prediction for COVID-19 using Faster R-CNN, Mask R-CNN, and ResNet-50,2021 International Conference on Advances in Electrical, Computing, Communication and Sustainable Technologies (ICAECT)
[5] He, P. (2020). Study on Epidemic Prevention and Control Strategy of COVID -19 Based on Personnel Flow Prediction. 2020 International Conference on Urban Engineering and Management Science (ICUEMS).
[6] Hua Ye; Peiliang Wu; Tianru Zhu, Diagnosing Coronavirus Disease 2019 (COVID-19):
Efficient Harris Hawks-Inspired Fuzzy K-Nearest Neighbor Prediction Methods, IEEE Access ( Volume: 9)
[7] Karthikeyan, C., Sunitha, G., Avanija, J., Madhavi, K.R., Madhan, E.S.,2021,Prediction of climate change using SVM and Naïve bayes machine learning algorithms,Turkish Journal of Computer and Mathematics Education.
[8] Kumar, N., & Susan, S. (2020). COVID-19 Pandemic Prediction using Time Series Forecasting Models. 2020 11th International Conference on Computing, Communication and Networking Technologies (ICCCNT).
[9] Nana Ramadijanti; Mu’arifin; Achmad Basuki ,Comparison of Covid-19 Cases in Indonesia and Other Countries for Prediction Models in Indonesia Using Optimization in SEIR Epidemic Models, 2020 International Conference on ICT for Smart Society (ICISS).
[10] Nanning Zheng; Shaoyi Du; Jianji Wang.Predicting COVID-19 in China Using Hybrid AI Model, IEEE Transactions on Cybernetics ( Volume: 50, Issue: 7, July 2020),Page(s): 2891 – 2904.
[11] Onder Tutsoy; Şule Çolak; Adem Polat,A Novel Parametric Model for the Prediction and Analysis of the COVID-19 Casualties, IEEE Access ( Volume: 8),Page(s): 193898 – 193906.
[12] Prajoy Podder M. Rubaiyat Hossain Mondal, Machine Learning to Predict COVID-19 and ICU Requirement, 2020 11th International Conference on Electrical and Computer Engineering (ICECE).
[13] Rachapudi, V., Varikuti, V., Anubrolu, J., Geethika, C.R.2017, A comparative analysis of classification algorithms for fetal growth,Journal of Advanced Research in Dynamical and Control Systems.
[14] Rajani Kumari; Sandeep Kumar; Ramesh Chandra Poonia; Vijander Singh,Analysis and predictions of spread, recovery, and death caused by COVID-19 in India, Big Data Mining and Analytics ( Volume: 4, Issue: 2, June 2021).
[15] Rajendra Prasad, K., Mohammed, M., Noorullah, R.M. 2019,Visual topic models for healthcare data clustering, Evolutionary Intelligence
[16] Ruizhi Han; Zhulin Liu; C. L. philip Chen; Lili Xu; Guangzhu Peng,Mortality prediction for COVID-19 patients via Broad Learning System, 2020 7th International Conference on Information, Cybernetics, and Computational Social Systems (ICCSS)
[17] Ryan Yixiang Wang; Tim Qinsong Guo; Leo Guanhua Li ,Predictions of COVID-19 Infection Severity Based on Co-associations between the SNPs of Co-morbid Diseases and COVID-19 through Machine Learning of Genetic Data, 2020 IEEE 8th International Conference on Computer Science and Network Technology (ICCSNT)
[18] Safa Bahri; Moetez Kdayem; Nesrine Zoghlami,Deep Learning for COVID-19 prediction, 2020 4th International Conference on Advanced Systems and Emergent Technologies (IC_ASET)
[19] Sina Ardabili; Amir Mosavi; Shahab S. Band; Annamaria R. Varkonyi-Koczy,Coronavirus Disease (COVID-19) Global Prediction Using Hybrid Artificial Intelligence Method of ANN Trained with Grey Wolf Optimizer, 2020 IEEE 3rd International Conference and Workshop in Óbuda on Electrical and Power Engineering (CANDO-EPE)
[20] Thirugnanasambandam, K., Prakash, S., Subramanian, V., Pothula, S., Thirumal, V.,2019, Reinforced cuckoo search algorithm-based multimodal optimization,Applied Intelligence [21] Yunxiang Liu; Yan Xiao,Analysis and Prediction of COVID-19 in Xinjiang Based on
Machine Learning, 2020 5th International Conference on Information Science, Computer Technology and Transportation (ISCTT).
