📊 COVID-19 prediction

This project aims to predict the result of three kinds of exams that detects the viral disease COVID-19: PCR, IgM and IgG tests.

It trains three feedforward neural networks with data from exams of the Albert Einstein Israelite Hospital, obtained from the COVID-19 Data Sharing/BR, available at https://repositoriodatasharingfapesp.uspdigital.usp.br/. The preprocessed data can be found at data/einstein.out.csv.

This is a project for the subject MAC0425 (artificial intelligence) at IME-USP. The final report (in Portuguese) can be found here.

Instalation

All the packages used can be installed with:

  pip install -r requirements.txt

Usage

The three models can be trained with:

  python3 main.py

Results

The models were validated by Cross-Validation with k-folds (k=10). The results are shown below.

Accuracies from Cross Validation

Repetition	PCR	IgM	IgG
1	0.644	0.568	0.639
2	0.715	0.542	0.604
3	0.701	0.600	0.500
4	0.500	0.570	0.613
5	0.738	0.653	0.670
6	0.704	0.542	0.549
7	0.723	0.649	0.500
8	0.682	0.640	0.678
9	0.736	0.625	0.610
10	0.716	0.566	0.577

Metric	PCR	IgM	IgG
mean	0.686	0.595	0.594
std. dev.	0.071	0.044	0.063
max	0.738	0.653	0.678
min	0.500	0.542	0.500

Confusion matrices

PCR

IgM

IgG

License

MIT