ML Guided Network Science Based Vulnerability Model

Introduction

The Covid-19 Pandemic has affected the whole world. As we slowly move towards recovering from this pandemic, we are bound to explore options that can help us effectively restore normalcy.
We use a LSTM based Time-Series prediction model which accurately predicts the future case counts. This in turn is used to calculate the Health, Social and Total Vulnerability (Cummulative Health and Social Vulnerability) of all the states of the USA.
We also calculate Travel bubbles or corridors (using the Total vulnerability calculated) that give an insight on which ones are the safest.

How to run

It is advised you use a virtual environment for running this model.

To create a virtual environment using pip, In the terminal/cmd type,

$ pip install virtualenv
$ virtualenv newenv
$ source newenv/bin/activate

To install the required packages or libraries for running the model type,

$ pip install -r requirements.txt

You would also need Rscripts to run the R files from Python.

Using the model

Important Functions

Here are the list of some important functions

Using the LSTM based Prediction Model

While initialising the LSTM_Model object, n_steps is used to specify the number of previous days' data the model will use to predict the next case counts. n_features is set to 1.

1. set_data(data_loc, data_type) is used to set the data for the prediction model to the specified database. The path of the database is stored in data_loc and the data_type can take values of the columns of the database ("new_case_count", "case_count", "death_count", etc.)

2. make_data(stateID) is used to split the data accquired from the function set_data() so that it can be fed into our Deep Learning model

3. model_compile() is used to compile the Deep Learning Model

4. model_train() is used to train the Deep Learning Model. The model is then saved with the filename CTST_2LSTM_100_<stateID>.h5 in the trained_models directory.

5. predict_3_days() is used to make prediction for the next three days.

6. update_csv() is used to firstly, make the prediction for the next day and update the important .csv files that will be used to calculate different vulnerabilities.

Using The Vulnerability Calculator

1. calculate_weights() is used to run all the R files. The R files perform a DEA analysis and calculate the health vulnerability of each state.
2. update_json() is used to update the json files that will be used to plot the necessary vulnerability maps and travel corridors/bubbles.

Plotting the results.

1. plot_state_social_vul() is used to plot the Social Vulnerabilities of all the states
2. plot_state_health_vul() is used to plot the Health Vulnerabilities of all the states
3. plot_state_total_vul() is used to plot the Total Vulnerabilities of all the states
4. plot_state_road_vul() is used to plot the travel corridors for the road networks. This also plots state centroids coloured according to their respective Total Vulnerability.
5. plot_air_transport_vul() is used to plot the travel corridors for the airplane networks. This also plots airports coloured according to their respective Total Vulnerability at airports.
6. plot_airport_vul() is used to plot airports coloured according to their respective Total Vulnerability.
7. plot_heatmaps(month) is used to plot heatmaps representing the vulnerability of different possible flights.

NOTE - 5, 6 and 7 do not update after predictions. (Future Work)

Code Implementation

Below is a code to plot the inter-state travel corridors on the road networks.

from models import LSTM_Model, Vulnerability_Calculator
from plot import plot_state_road_vul

lstm_model = LSTM_Model(n_steps=7, n_features=1)
# Predict next day value and update the required csv
lstm_model.update_csv()

cal = Vulnerability_Calculator()

# Using R codes calculate files and update the required json files
cal.calculate_weights()
cal.update_json()

# Plotting Road Vulnerabilties
plot_state_road_vul()

Output

The output of the above code is,

Other Outputs

The social vulnerabilities of the state can also be calculated. Shown below is one of the results that we get,

A heatmap for the month of October is generated to demonstrate the severity of flight travels across different airports,