This study sought to apply the SVR technique to predict the ICU bed occupancy rate by COVID-19 in Brazil for 7, 14, 21 and 28 days after May 10, 2021. See the document here.
-
Data Source: https://bigdata-covid19.icict.fiocruz.br/
-
Number of instances: 27 instances
-
Date of last instance: 10/05/2021 (day / month / year)
-
Final data:
- Training data: 85%
- Validation data: 15%
- Metric: MAE (Mean Absolute Error)
- No. of training and validation tests: 10
- Prediction time intervals: 7, 14, 21 and 28 days after the last collection date
Figures 1, 2, 3, 4, 5, 6, 7, 8, 9 and 10 show the graph referring to the training and validation in each testing testing phase, respectively.
The table shows the parameters that were changed in each test, as well as the respective MAE results. The parameters gamma and coef0 were constant for all tests, with the values 'auto' and 1, respectively.
In the testing phase, the parameters used in the 6th test were chosen because of the lowest MAE value obtained (8.80%), so the SVR function was as follows:
SVR(kernel='poly', C=1, gamma='auto', degree=8, epsilon=0.1, coef0=1)
The figure below shows the result of this prediction:
- red dots: actual occupancy rates already available in the dataset;
- blue curve: regression for the already known values
- red curve: prediction of future days
SÁ, Gabriel Caldas Barros e et al. (2021) Predição Da Taxa de Ocupação de Leitos de UTI Por COVID-19 No Brasil Usando SVR.. In: Anais do Congresso Brasileiro Interdisciplinar em Ciência e Tecnologia. Anais...Diamantina(MG) UFVJM.