Inspiration

Inspired by the situation in Delhi, India, wherein people who get the report of "COVID-positive" are expected to visit a hospital and queue up for hours together for a chance to get a bed at the hospital. If they do not get a bed, they go to another hospital, queue up for hours, and hope for a bed there too. This movement of COVID-19 positive patients from hospital to hospital and queueing up is troublesome in two of many ways. Firstly, this movement increases the risk of spreading of the virus itself as these patients visit many different hospitals in the hope to get a bed, thereby infecting so many different places. Secondly, these patients do not get immediate care, and are forced to spend time waiting in queues. An added problem arises in beds being "booked" by people in power and those that can pay the highest for that bed. A news report featured a shocking amount asked by a COVID-19 patient of 1 lakh INR (£1061) per bed per night. During the current times of pandemic, corruption in healthcare must be condemned. Beds must be allotted based on case severity and not on the amount of money.
This inspired us to make an application that would help overcome these issues, and be able to allot a bed to the person, most in need, without them having to queue up and go hospital-to-hospital.

What it does

What if we could automate this system of "queueing-up" for a bed? What if we could have a system wherein COVID-19 positive patients would know exactly which hospital to go for a guaranteed bed, at the ease of their phone?

Keeping this in mind, we created a mobile application that lets:
1) COVID-19 patients look at bed availability in hospitals nearby,
2) COVID-19 patients send one e-application (through the app) for a bed to hospitals within 10km radius, by filling in details (such as age, past illnesses, travel history etc.) and attaching COVID-19 positive report scan,
3) The Machine Learning algorithm predicts and ranks the severity of the case/e-application using these details per bed space,
4) Based on the results of the ML model, hospital officials can be suggested on who the bed should be allotted to instead of allowing officials to indulge in allotting the bed to the highest bidder (the one with most money)
5) A hospital then allots an available bed to the most critical case/application after which the case/application is marked as "Treated" on the database of applications, which would signal to other hospitals that this particular case has already been catered to
6) The person who has been allotted the bed, gets a notification and a set time to arrive at.

In other words, we are creating a database of beds and the applications for those bed spaces, which are then ranked on the basis of severity using questions answered by the patient. The most severely ranked case gets a bed. This process brings about transparency and helps cater to those who need it most. It would aid in removing any corruption that exists in allotting bed spaces.

How we built it

The steps we followed for creating the ML model are:
1) Acquiring the Dataset: We found an appropriate dataset for this particular use case, on Kaggle.
2) Data Visualisation: We tried to visualise the data using various techniques and plots. We built correlation matrices, scatter and density plots, histograms to see which symptoms affected severity the most.
3) Data split: We split the data into 70:30 train:test.
4) Training: We trained the random forest classifier model on 70% of the data.
5) Testing: We then tested the model using the rest of the data.
6) Evaluation: We used various evaluation metrics to judge the performance of the model. It possessed a mean score of roughly 74%.

We then designed the app 'Covey', keeping in mind a simple user interface allowing easy input of data from the user point of view.

Further, we included a code written in JavaScript, which involved the Twilio framework to develop a messaging system to notify a patient if they have been allotted a bed at a hospital.

Challenges we ran into

Finding an appropriate dataset was the hardest task. This is because of the limited research on the ongoing pandemic. Secondly, downloading a few libraries that would satisfy the version requirement, took a significant bit of time and debugging.

What's next for COVID-19 Automatic Bed Allotter

We wish to increase the accuracy of our ML classifier and train it on a more intensive dataset, which includes more combinations of symptoms and other illnesses.
We are hoping to completely integrate the ML model with the app. Also, we hope to automate the notification being sent after a bed is allotted, as currently it requires a prompt to run the code which then triggers a notification.

Try out our App!

https://xd.adobe.com/view/3dbb8b3d-33da-4aca-9b7e-68ef428f888e-68e8/

Built With

Share this project:

Updates