Following the theme of this hackathon, our project focuses on time series analysis and forecasting of air quality. As we know, post lockdown due to coronavirus, we have come across many images and memes on the internet talking about a drastic improvement in the environment with respect to pollution and we want to demonstrate this through numbers and graphs. Our aim here is to show how the pace of our daily lifestyle is impacting our environment and spread awareness about how we should start to change it which is very easy to overlook amidst our busy lifestyle.

What it does

The dashboard we have built shows the time series forecast for the next 6 months from June to November for New Delhi. We have air quality measurements in terms of 6 parameters. From the plots in the dashboard, we see how the lockdown is impacting the environment for the better. We show the graphs only for New Delhi but it can easily be generalized to all across the world as the dataset is easily accessible via OpenAQ dataset extracted on Amazon Athena.

How we built it

We extracted the air quality dataset via Amazon Athena from OpenAQ source. It starts from May 2015 and is updated daily. We have averaged it over a month and have used monthly mean for the forecasting. We have used LSTM based neural network model for the same and built a dashboard using HTML, CSS and Javascript to demonstrate our findings. Ideally, all the parameters except O3 should be low. The true values for the months of March and April in New Delhi, which was the lockdown period see the expected fall in their values and O3 sees a rise. We also see the predicted values by our model which shows slightly higher values as it sees these three readings as outliers.

Challenges we ran into

Time series forecasting is an entirely new concept that we learnt and built the model from scratch. We also had a lot on our plate to build an end to end project.

Accomplishments that we're proud of

We were able to implement most of whatever we had in mind and have a complete project with a working UI, backend and a Machine Learning model.

What we learned

Time series forecasting, deployment to Google App Engine, Domain registration

What's next for

Expand it to other regions as well and build a more interactive UI.

Share this project: