leghart/covid-analyzer

Table of contents

• General info
• Packages
• Setup
• Prediction
• Periodic data download

General info

A project aimed at collecting up-to-date data on the coronavirus pandemic and predicting new cases for the next few days. Additionally, the data collected with the prediction are automatically placed on the website. In future page will be probably developed to make easy access from anywhere.

Packages

Project has created with:

• BeautifulSoup
• SqlAlchemy
• Tensorflow
• Pmdarima
• Flask
• SMTPlib

Setup

All .py files must be in one directory. Depending on the path to the directory, change the .bat file (this is described in "Periodic data download"). To run the program, run the main.py file which is responsible for downloading data and uploading them to the database. The page.py file is responsible for launching the website. Before used code, you have to prepare setup.py file. There are a few important fields:

• Country - choose a country to scrap data, insert to database etc.
• scrap_time - different countries have different upload-data times.
• forecast_hor - how many days you want to predict in future (too high value will make the network useless)
• db_name - name of the database to be created and connected in the future where the data will be stored

Prediction

The following prediction functions have been implemented in this project:

• ARIMA - Autoregressive integrated moving average is trying to forecast a future cases. ARIMA models are applied in some cases where data show evidence of non-stationarity in the sense of mean. Actually the best solution which is used. Implemented from pmdarima.
• LSTM - Short-term memory neural network with tensorflow. The network was taught with 75% of the collected data (the remaining 25% is the test kit). The learned model can be used as simple predictor, but it has a large error, because we don't have enough data for good learning and new cases depend on things, which I do not collect (e.g. lockdowns information, possible virus migration, temperature etc). It's just a simple project to get to know neural networks better).

Periodic data download

Currently I'm working on making full-automatically system to download data based on Raspberry.