jairojuunior/covid19-prediction-labtests

Analysis and prediction of SARS-CoV-2 based on laboratory tests

Analysis and prediction of SARS-CoV-2 based on laboratory tests

Kaggle competition: https://www.kaggle.com/dataset/e626783d4672f182e7870b1bbe75fae66bdfb232289da0a61f08c2ceb01cab01/data

We analyse feature bias through severity of cases. 6 groups of tests were found using Principal Components Analysis. 3 of them are: Regular blood test, respiratory virus and bacterial tests and Influenza rapid tests). For these we fit models and compare the ones that favors precision and recall through a clinical perspective. Regular blood tests were found to be the most useful to detect coronavirus. We also backtest our best models agains patients infected with H1N1, Rhinovirus and Influenza B and found that our models capture SARS-CoV-2 patterns different than these flus.

Our results are summarized in the presentation file (PT-BR) COVID-19_ Influência de exames na precisão e recall de modelos.pdf.

Please find covid-19-influence-of-exams-in-recall-precision.ipynb for in-depth understanding of methodology employed in this work.