For latest predictions, please visit our tracker.
Currently we are seeing that Mumbai, Delhi, Chennai and Ahmedabad together account for 2/3rd of the cases. But the same trend would not continue for long. When all of this is over, UP would end up being the state that would report the maximum number of daily cases.
A peak date is when the epidemic growth is at the maximum and after that the disease starts to subside. Hence it is of interest to know if a country or state has peaked. In our previous post, we described how we trained a 20 day forecaster using deep learning and a Recurrent Neural Network. Feeding back the predicted data as input data in a recursive loop, we can extend the prediction even further. We did exactly this and extended the forecast for next 6 months. We found that the following states would end up reporting more than 10K cases daily:
Then these states would report more than 5K cases daily:
Lastly, these states would report more than 2K cases every day:
Now summing up all the state level graphs, we calculate the aggregate graph for India which looks like this:
Peak date: Oct 6, 2020
Peak daily new cases: 93,414
Cumulative confirmed cases: 1,14,46,050
A more textured graph can be drawn once we have district wise granular data.
Our calculations and code is available in this notebook.
Last Updated: June 11, 2020.
We plan to update these results every few days. So please check back for any changes.
|State||Peak date||Peak daily cases||Cumulative cases|
|Andaman & Nicobar Islands||Oct 01||69||3,545|
|Andhra Pradesh||Sep 20||14,311||6,97,062|
|Arunachal Pradesh||Nov 08||3,558||2,28,398|
|Dadra & Nagar Haveli and Daman & Diu||Nov 24||106||4,297|
|Himachal Pradesh||Sep 27||766||58,292|
|Jammu & Kashmir||Aug 09||2,151||1,10,561|
|Madhya Pradesh||Oct 21||8,837||7,19,312|
|Tamil Nadu||Jul 17||11,386||6,29,946|
|Uttar Pradesh||Oct 01||27,400||17,98,338|
|West Bengal||Aug 31||8,665||5,58,697|
States in the order they peak, with peak daily cases listed as bars:
The graphs for a few states were predicted incorrectly by our RNN because of insufficient data. The secondary wave has to be ignored for now till we refit the model with more training data.