Epidemic Prediction

Term project for ENSC413. An application of deep learning to help predict where an epidemic will occur next.

Installing the Software

If you haven't already, clone the repo as follows:

git clone git@github.com:brittanyhewitson/epidemic_prediction.git
cd epidemic_prediction

Or use https:

git clone https://github.com/brittanyhewitson/epidemic_prediction.git
cd epidemic_prediction

Next, create an environment. You can use conda as follows:

conda create -n epidemic_prediction python=3.5
conda activate epidemic_prediction

Alternatively, you can use venv:

python3 -m venv venv
source venv/bin/activate

Now install the requirements for the software:

pip install -r requirements.txt

Finally, run the setup file:

python3 setup.py develop

Testing the Models with Various Parameters

To test the models with a variety of parameters, the test_parameters.py script can be used as follows:

python3 test_parameters.py --gpu --data_choice small_data

Where the --gpu flag indicates whether you wish to use a GPU and the --data_choice option specifies which data source to train on. The choices for data source are:

Data Source	Description	Size
big_data	The complete raw input dataset. This file is large and imbalanced, therefore it will take several hours to train and will produce results of approximately 87% regardless of the parameters	116109 rows
small_data	The small input dataset provided by the example article. This dataset is found in the X.pkl and y.pkl files and is relatively small compared to the large dataset. Therefore it will take approximately 5 minutes to run and should produce accuracies in the range of 80-95%	1213 rows
smote	The complete input dataset upsampled using SMOTE. This dataset includes all rows from the input dataset, therefore holds data for all dates for all locations	201948 rows
by_date	A number of CSV files, each representing the feature and output data for all locations on a single date. Only to be used to find the accuracy as a function of the date	26 CSV files
single_date	Uses only a single CSV file from the list of CSV files in the by_date category	Size depends on which date is selected, varies from approximately 1000-2000 rows per file

The default value for the data choice is a single date, with the date beign 2016-06-25.

Running test_parameters.py will generate and save a series of plots relating to the accuracies produced by varying parameters for each model.

Testing the Models with Optimal Parameters

To test the models with optimal parameters, simply run the following:

python3 run.py

This will train and test the perceptron, multi-layer perceptron, SVM, k-NN, and random forest according to the parameters we have found that produce the best results. The accuracy for each classifier will be printed to the console, and a bar plot comparing the accuracies will be displayed and saved in a new directory called output_images.