AHMEDABAD: The third wave of Covid-19 may affect India between July 15 and October 13, with a peak in the daily number of cases expected in the first two weeks of September.
These are the forecasts of a recent study conducted by a Pandit Deendayal Energy University (PDEU) professor along with two Nirma University alumni.
The study titled ‘Pattern Recognition: Prediction of Covid third wave in India using time series forecasting with deep learning models’ has been submitted to Elsevier, a scientific journal.
The study takes into consideration the vaccination numbers till the second week of June.
“If the government’s plans of increasing the vaccination rates five-fold over the next few months are successful, the peak in the third wave will be about 25% of the peak Covid-19 cases in the second wave,” said Manan Shah, assistant professor, department of chemical engineering at PDEU, who was part of the research team.
Shah added: “So, it could be somewhere around 1 lakh Covid-19 cases per day.” The study focuses on two likely scenarios of the third wave.
The government’s mega vaccination drive that is currently underway is expected to make a big difference of about 85% as compared to the vaccination numbers a month ago.
According to one model, the peak number of cases will reach more than 6 lakh per day in September if the vaccination rate remains what it was till mid-June, without any major changes in it.
The second wave peaked at around 4.20 lakh daily cases in May.
“As per our study, we have predicted a scenario where the number of daily Covid-19 cases will reach a peak of 6.36 lakh around September 25,” said Atharva Shah, an alumnus of Nirma University.
“While the number of cases will peak in the first two weeks of September, they will be recorded around the last week of September.” Maharishi Gor, also an alumnus of Nirma University, is another co-author of the study.
Using machine learning and AI techniques, the study proposes a deep learning architecture that can forecast the number of cases with vaccination data.
“The dataset provided by Johns Hopkins University is taken and the proposed methodology is applied,” according to the research paper.
“The model is trained on the US data and the Indian data is used as a test dataset.
The model performs well on both the datasets with good forecasting accuracy and reasonable error metrics.” The paper adds: “The model captures the trend of the virus spread and is visualized graphically.” Covid-19 has led to infrastructure collapse in many countries such as India, the paper says.
In the vulnerable cities, the paper says, the population density and unstructured layout have made the cases explode with over 29.3 million cases to date with most requiring medical attention.
Two inputs were considered for the study — the number of new cases and the vaccination percentage.
The US and India were chosen for the study as they were the two worst affected countries in the pandemic.
The US was chosen as a training set while India was used as a test set and as a forecasting space.
“As vaccination is speeding up across most of the developed countries, India is still at a 3.2% vaccination count as compared to the US which stands at 41%,” according to the paper.
“Thus, it becomes important to predict the rise of cases to aid the government in understanding the allocation of resources and arrest the spread of this deadly virus.” The paper is currently under review at Elsevier for publication.
“We hope it will be published soon,” said Shah of PDEU.