Authors

Inspiration To help understand the growth of COVID cases in future which would enable countries to take proactive measures such as necessary supplies, and infrastructure, to cope up with the new upcoming cases

What it does: It forecasts the worldwide COVID (confirmed, deaths and recovered) cases for next 30 days starting from 28th March

How I built it: I used python (Keras) where I trained and tested the time series LSTM model using the COVID data received from 22nd Jan to 28th March 2020, after splitting the data into train-test cases. The predictions are done for each set of confirmed, deaths and recovered cases of data. The source of input data is https://coronavirus.jhu.edu/map.html

Challenges I ran into: The challenge was to validate or test the trained LSTM model. However, after varying the number of neurons in a hidden layer, number of iterations, and time-steps, which was time-consuming, I was able to validate the model.The challenge was also to perform the future predictions once you trained the model, because the complex LSTM model requires reshaping the data in a certain format before implementing the model.

Accomplishments that I'm proud of: Main accomplishment is to be able to test the model with greater than 80-90% accuracy, and finally perform the future predictions. Other accomplishment is that I am able to use LSTM time series neural network, which can store long past sequences of memories and thus is able to do forecasting further long into the future.

What I learned: I learnt implementing LSTM model for training and future predictions. It requires tuning few LSTM parameters before model can be trained which is a time consuming process. Further, I gained insight on ways to do future predictions after formatting the data, which was a huge challenge,

What's next for Forecasting of worldwide COVID cases for next 30 days: Next step is to forecast the COVID cases (confirmed, recovered and death) for specific countries across world (including USA and European countries, where crisis is more severe), so we can gain more insights on future growth/decline patterns country-by-country.

Try It out

Hackathons

Technologies

keras, matplotlib, numpy, pandas, python

Devpost Software Identifier

257700