Predicting Covid-19 by Using a Machine Learning Algorithm

The SARS-CoV-2 epidemic has caused disaster all across the world. Machine Learning (ML) and Cloud Computing may be used to track the disease, anticipate epidemic growth, and devise strategies and policies to control its spread. An enhanced approach based on machine learning has been used to estimate the possible danger of COVID-19 in nations worldwide. This has been placed on a cloud computing platform for more accurate and real-time growth prediction.

Introduction

On December 31, 2019, the new Coronavirus illness (COVID-19) was first reported in Wuhan, Hubei Province, China. The cumulative incidence of the causal virus (SARS-CoV-2) is fast growing and has touched 196 nations and territories, the most afflicted being the United States, Spain, Italy, the United Kingdom, and France. In the absence of a cure, the only option is to restrict the spread of the virus by using "social distance" to break the infection's chain of transmission.

Prediction Model and Performance Comparison 

The novel coronavirus has a significant socioeconomic impact throughout the world. As COVID-19 develops in popularity worldwide, more data must be collected, developed, and analyzed. To better understand how the global population is impacted, we propose a Machine Learning model that can be run constantly on Cloud Data Centres (CDCs) for accurate spread prediction and proactive strategic response creation.

Machine Learning Model

Many studies have found that data relating to new instances over time has many outliers and may or may not follow a conventional distribution such as Gaussian or Exponential. The data from an earlier version of the virus, SARA-CoV-1, matched the Generalized Inverse Weibull (GIW) Distribution better than the Gaussian distribution. However, in this work, the regression curves were constructed using the Susceptible-Infected-Recovered model, and the number of cases was estimated using a Gaussian distribution.

Distribution Model Selection 

Analyzed data on daily new confirmed COVID cases to determine the best fitting distribution model for the data matching COVID-19. Five sets of global data were employed to fit the parameters of various distributions. Compared to iterative versions of Gaussian, Beta (4-parameter), Fisher-Tippet (Extreme Value distribution), and Log-Normal, the Inverse Weibull function best fits the data. When applied to the same dataset, Iterative Weibun had a 12 % lower average MAPE than non-iteratively weighted Weibul.

Evaluation of the performance of the algorithms 

The data were subjected to 10-fold cross-validation, in which the data were randomly divided into training and validation cohorts at an 80/20 ratio ten times. Because our data was unbalanced (only 2.1% of the output had the condition vs 97.9 % without), we used an oversampling approach to improve learning on the training data. The algorithms' performance was assessed regarding discrimination, calibration, and overall performance. We investigated discrimination using the area under the receiver operating characteristic curve (AUC) and calibration using accuracy and the Matthews correlation coefficient.

Biases in data

Different nations have responded differently to the SARS-CoV-2 epidemic and its associated illnesses, COVID-19. The suggested GIW model is applied to each country independently to match the model parameters to the distribution of new instances over time. It combines biases from citizens' and migrants' travel history and lockdowns and social distancing tactics implemented by each nation.

Leveraging tracking systems for near-real-time predictions 

The transmission of the illness may be tracked using effective and up-to-date tracking techniques. Only by thorough and organized testing will we limit the harmful impacts of the disease's spread. Government organizations can use cloud services to create such frameworks, inputting data from such tracking devices and predicting the number of instances shortly in near-real-time. This allows the authorities to gradually release the lockdown, maintaining a cover on the post-lockdown increase in cases.

Beyond the lock-downs 

Travel and group activities are currently prohibited all across the world. When the lockdowns are removed, the number of new illnesses and deaths may deviate considerably from the expected patterns. Other variables, such as viral mutations, will impact the spread in the future. As a result, the ongoing effort is necessary to ensure accurate projection and appropriate safeguards are implemented.

Research Opportunities and Emerging Trends

The COVID-19 pandemic has opened up numerous new study avenues for current and future pandemics. The following are some of the most outstanding research opportunities.

1. Incorporating other indicators: To improve prediction accuracy, essential parameters such as population density, age distribution, separable and community movements, level of healthcare accommodations available, strain type and virulence of the virus, and so on, must be included in the regression model.

2. Integrating with other time series models: ARIMA models may be combined with the Weibull function for further time series analysis and prediction.

3. AI may be used to anticipate the structure and function of proteins associated with CoV-2 and their interactions with human host proteins and the cellular environment. By creating appropriate algorithms, the contribution of different socioeconomic elements that impact the epidemic's vulnerability, spread, and development may be anticipated. This can aid in the effective distribution of resources in big nations with limited healthcare resources.

4. Using AI to evaluate social media data: We can also examine and analyze social media data for real-time gathering COVID-19 epidemiological data.

5. Robotics for contactless treatment and medication delivery: AI-powered robots can be utilized to do contactless delivery and treat patients remotely, reducing the engagement of medical personnel with infected individuals. Furthermore, COVID-19-enforced lock-downs have resulted in significant improvements in global air quality.

6. Climate Change: Due to COVID19-enforced lock-downs, air quality has improved significantly worldwide. However, there is widespread speculation of retaliatory pollution due to these lockdowns. Future research may include more thorough investigations of age distributions and demographics with other features.

7. Risk assessment: Using AI, the risk of serious illness associated with COVID-19 for persons of various ages may be predicted. Proactive steps can be done using such algorithms to avoid virus transmission to vulnerable segments of society.

8. Real-time sensors and visual imaging: AI-based preventative steps can be implemented to prevent the virus from spreading to vulnerable societal sections.

Real-time sensors, such as those used in traffic cameras or surveillance, can track COVID-19 symptoms using visual imaging and tracking applications and notify hospitals and administrative authorities for disciplinary action. Tracking must cover all steps, from entry points to public areas and hospitals.

References

1. Tuli, Shreshth, et al. "Predicting the growth and trend of COVID-19 pandemic using machine learning and cloud computing." Internet of Things 11 (2020): 100222.

2. DAS, ASHIS, Shiba Mishra, and Saji Saraswathy Gopalan. "Predicting community mortality risk due to CoVID-19 using machine learning and development of a prediction tool." medRxiv (2020).