NYC Covid-19 resource allocation prediction application

This project was the capstone for the computer science Bachelors degree program I completed in 2021.

The requirements of the project were:

1. Implement one descriptive method and one non-descriptive (predictive or prescriptive) method.
2. Use of collected or available datasets.
3. Includes a decision-support functionality related to the non-descriptive method.
4. Use methods and algorithms supporting data exploration and preparation to make data useable.
5. Implementation of interactive queries to access specific data.
6. Include a machine-learning method(s) and algorithm(s) to support the decision making functionality.
7. Incorporate functionalities that will evaluate the accuracy of the application’s outcome.
8. Provide an industry-appropriate security features that protects access to the application of data.
9. Include a tool(s) to monitor the applications functionality for maintenance purposes.
10. Include a user-friendly, functional dashboard that includes at least three visualization types (not just images) for data exploration and inspection.

This project could be a standalone program, or a web or mobile application. I chose to create a web application using the NextJS framework, hosted on Vercel.

The backend functionality was written in Python using Flask for a lightweight server, and Tensorflow for the machine learning model creation and training.

The dataset came from the publicly-available data api provided by https://data.cityofnewyork.us.

Retrospective

Looking back now, there are several things I would have changed or added for this project:

1. Saving the TF model after training so it doesn't need to happen on every request. This takes upwards of 2-3 minutes, which tends to appear as though the request has failed.
2. Before learning about the above solution for saving & reusing models, I thought about streaming status updates of the model creation/training process on the server, either using WebSockets or Streams.