COVID-19 has disrupted all our lives. More and more people are getting infected at more and more places and this has resulted in people dying, economic hurdles and disruption of everyday activities. One of the major problems that this has resulted in is over flowing of hospitals in a few areas. Due to the limited supply of essential medical supplies like Ventilators and Beds, people are not able to get the medical service they need to survive. This got us into thinking, What can we do about this issue? To address this issue, we developed a model that first predicts the number of cases and then calculates the number of medical resources required based on this prediction.

What it does

Our Project predicts the number of COVID-19 cases that might occur in a given area and then uses this prediction to calculate the number of medical resources that are will need. All of this is Data and Analysis is then displayed onto a Web Interface. Through this web interface we have enabled a notification system that people (including Medical Professionals) can choose to subscribe to via which they get a report on the current situation at the specific area of their choice.

How we built it

We built our project with a number of goals in mind. Our primary goal was to build something that would be helpful to the nurses and doctors at hospitals across the country in preparing for the task they have lying ahead. We decided a web application that displays a map with markers of data and projected predictions would be the best way for healthcare professionals to visually understand what is happening. We used a combination of HTML, JavaScript, and CSS to design the webpage so it is user friendly. To acquire the data, we decided to develop a machine learning model that is able to use past data to project data for the upcoming days. Keras API and Tensorflow were used to develop and train our model at 200 epochs. The data we collected, which is accessed by both the web application as well as the model, is stored in the cloud using MongoDB Atlas. This allows a centralized place for data storage and makes it easy to add additional data as it is gathered.

Challenges we ran into

Some of the major challenges that we ran into came with deep learning. Deep learning itself is a relatively new and dynamic sector of computer science: there's always more to learn and experiment with. As a result, we tried to find a model that would work well with our problem. We needed a model that could use a time-series with multivariate data to predict the number of covid19 cases per US state. This ended up taking a majority of our Saturday, as none of our team members had much experience with LSTM and deep learning jargon. We explored Vector AutoRegression (VAR) and Multilayer Perceptron Models, and after major testing and technical difficulties like our laptops crashing mid-run, we were able to find a model that actually predicted with our testing data with some accuracy.

Accomplishments that we're proud of

In this project, we were successful in achieving our goal of developing a web application that notifies healthcare professionals of the predicted upcoming situation of the Covid-19 pandemic. Doctors and nurses will be able to use our project to be better prepared for the task they have ahead of themselves. Our machine learning model was accurate to a very high degree, so the models and actual data are as close as possible. We were able to obtain a validation RMSE of 33 and a testing RMSE of 42. We were able to successfully implement a cloud-based database on MongoDB Atlas for ease of future updates. The centralized data also makes it easier for the model and the web application to both access the same database, allowing the future of this project to remain bright. Lastly, we succeeded in creating a visual display to overlay on a map so the data can be easily understood using Google Maps API.

What we learned

More than just learning the importance of collaboration, our team was able to explore several new technologies and ideas. One of the difficult tasks we learned from was developing a time-series deep learning model through Keras Deep Learning API. We spent hours getting familiar with some of the mathematics and conceptual ideas (such as one-hot-encoding), which helped open our eyes to the countless applications of computer science and deep learning, especially in the health sector. In addition, we got familiar with the MongoDB NOSQL database, as we built a database and used the data from it to train our deep learning model.

What's next for COVID Predictor & Notifier

We aim to find more Datasets that help in increasing the precision of our models. Next, Depending on the Datasets we find our project will identify and predict more medical supplies categories like masks, sanitizers etc. This would be followed by enabling a push notification system to mobile phones via SMS.

Try It out



amazon-web-services, css, google-maps, html, javascript, keras, mongodb, node.js, python, tensorflow, twilio

Devpost Software Identifier