AI Model Performance: Predicted vs Observed Cases for Coahoma, Mississippi
AI Model Performance: Predicted vs Observed Cases for Charles City, Virginia
AI Model Performance: Predicted vs Observed Cases for Blount, Tennessee
AI Model Performance: Predicted vs Observed Cases for Bosnia and Herzgovnia
Example of Storage of Telediagnosis Form in Firebase
Example of Storage of Violation Report in Firebase
We are in the midst of a global pandemic that is showing no signs of slowing down. As we all continue to play our part in social distancing, we want to provide the public with a web application that disseminates and collects data regarding current pandemics, their spread, and their impact on our world.
What it does
Safedemic allows users to report violations of pandemic-related policies, visualizes the current and predicted spread of pandemics via heat maps, and provides health guidelines, live news, and live statistics to inform users about recent developments regarding the specified pandemics. Safedemic will also ease the process of telediagnosis by allowing users to communicate their medical status to doctors through report forms.
How we built it
- Integration with Firebase Database
- Login Screen
- Report a Violation Form
- Creation of current spread heatmaps
- Creation of violation report map markers
- Pandemic news panel
- Obtain live data for total cases and country-based statistics
- Include certified health guidelines
- Include option to view the predicted spread in a user-defined location within a user-defined circular region (see machine learning workflow for further details)
- Include telediagnosis form which can be sent to doctors (via email) for review
MACHINE LEARNING WORKFLOW:
- Compile pandemic data from John Hopkins
- Split data into train and test sets
- Create a multilayer perceptron to predict the number of cases x days into the future (where x is an input)
- Generalize the model to be able to break down the predicted new cases by location as well (in order to visualize predicted spread on map)
- Train the model on the training set
- Assess the performance of the machine learning model at each individual location and as a whole
Challenges we ran into
For the heatmaps pertaining to current spread, it became largely time-inefficient to obtain live data using web requests. Thus, we decided to store the data into a local .csv file and update the data every 10 hours. For the machine learning model, we had issues linking the model with the prediction spread visualizations on the map. Consequently, we stored the model predictions into a local .csv file and updated the predictions every 10 hours.
Accomplishments that we're proud of
We were proud that we could provide a more efficient form of telediagnosis (as opposed to time-consuming video conferences) and allow civilians to share/report violations of pandemic-related guidelines and laws. Furthermore, we are satisfied with how we visualized the violation reports, current spread, and predicted spread in the map section of the website.
What we learned
We learned how to effectively use the Google Maps API and News API. However, what was truly interesting about our endeavors was how we were able to integrate our prior machine learning and web development experience in order to create this web application.
What's next for Safedemic
We plan to include prediction spread visualization for all countries across the world. Moreover, we plan to attract users initially by using social media outreach on Facebook, Instagram, and Twitter. In addition, we will contact our local law enforcement officers and state representatives in order for them to advertise our product to their larger network. We believe given the current situation that many people will download our application in order to report violations as it is in their personal interest. Ultimately, we plan to attract users nationwide along with targeted advertising in areas with high numbers of infected people and metropolitan areas (NYC, London).