/COVID-19-Trend-Prediction-using-Recurrent-Neural-Network

Predict whether the number of confirmed people will increase or not with the use of RNN.

Primary LanguagePythonMIT LicenseMIT

COVID-19-Trend-Prediction-using-Recurrent-Neural-Network

Predict whether the number of confirmed cases will increase or not with the use of RNN, LSTM & GRU.

Dataset

Observe the sequences of comfirmed cases during 2020-1-22 ~ 2020-4-12 from 185 countries.
The gloabl data of comfired cases please refer to 👉 https://ourworldindata.org/coronavirus

image

Execution & Overall Structure of system

  1. Sequence Preprocessing : find high correlated countries & prepared segments for sequence modelling 
    python3 Preprocess.py
  2. RNN for trend prediction : training the model with Torch package. 
    python3 RNN.py
  3. Visualization on a world map by "pygal" packag 
    python3 WorldMap.py

Sequence Preprocessing

  1. Compute to correlation coefficient (CC) between 2 countries.
    image

  2. Add the pair of countries to set C if their CC higher than 0.7 (highly-correlated)
    image

  3. Generate the data pair (segment,lebel) for modelling from set C

RNN for Classification

  • RNN with 2 layers & dropout
    In this project, dropout is adopted to aviod overfitting problem.

    image

    image

Visualization of prediction on a World Map

image