ML Model Development to Predict ICU Admission for Confirmed COVID-19 Cases in Brazil

Brazil is one of the countries most affected by the COVID-19 pandemic, with more than 16 million confirmed

cases and 454 429 confirmed deaths by May 26, 2021 (according to the Johns Hopkins Coronavirus Resource Center). 

Brazil was and still is one of the countries most impacted by the first wave of Covid-19 which first recorded case on 26th February 2020 and reached community transmission from 20th March 2019 onwards to dates, that caught Brazil unprepared and unable to response due to the strain on hospital capacity such as the intense and lengthy request for ICU (incentive care unit) beds, professionals, personal protection equipment and healthcare resources.

A data science team at Sirio Libanês, a top-tier hospital in Brazil, decided to use ML to help reduce the strain on hospital’s ICU beds where the objective is to develop a ML model to predict if a patient of confirmed COVID-19 case would require admission to the ICU.  With that objective in mind, the team has collected a decent amount of clinical data from patients i.e. the features of COVID patients, and the target (those been admitted to ICU).

The dataset is released on Kaggle platform with full data description at the following URL (https://www.kaggle.com/S%C3%ADrio-Libanes/covid19) by the team seeking interesting solutions and findings from the public.

In this assignment, you are challenged to perform a full lifecycle ML model development according to the objective of the dataset, which includes the following elements:

1) Perform exploratory data analysis (EDA), and establish hypotheses of predictive insights you expect to glean from the dataset.

2) Perform data preparation for ML informed by the EDA findings.

3) Develop ML model according to hypotheses of predictive insights you gleaned from the dataset. You are required to evaluate at least 3 ML algorithms and assess associated issues i.e. hyperparameters tuning, performance metrics, model complexity (underfitting/overfitting) etc.  Finally, provide a recommendation of the best algorithm for your ML model.

4) Documentation:

i) Jupyter-Notebook include all coding and technical report i.e. explanations, justifications, reasonings etc. for every finding, strategical decision, action, and choice made.  Jupyter-notebook provide a comprehensive documentation capability by using the Markdown (https://www.datacamp.com/community/tutorials/markdown-in-jupyter-notebook).

ii) Executive Summary Report (maximum 1000 words) to provide an overview of the entire lifecycle of the ML model development, written with target audience in mind such as high-level stakeholders, decision makers, directors etc.