COVID-19 Death Risk Prediction Analysis

Project Overview

This data analytics project focuses on developing a solution to predict the risk of death from COVID-19 using a dataset provided by the Centers for Disease Control and Prevention (CDC). The CDC dataset includes de-identified individual-case data, capturing various aspects such as demographic information, exposure history, disease severity indicators, clinical and laboratory diagnostic test results, co-morbidities, and survival outcomes. The goal is to leverage this data to build predictive models that can accurately determine the likelihood of death ("death_yn") based on descriptive features.

Dataset

The dataset used in this analysis is sampled from the public data released by the CDC, which tracks cases, deaths, and trends of COVID-19 in the United States. It consists of standardized case reporting forms submitted on a daily basis. For this project, we focus on using this dataset to build and evaluate prediction models for death risk based on various factors.

Methodology

The project is structured into several key sections, each aimed at understanding the data, selecting promising features, and developing predictive models:

1. Data Understanding and Preparation: We begin by splitting the dataset into a 70% training set and a 30% test set. This phase involves exploring the relationships between feature pairs and selecting promising features for modeling.

2. Predictive Modeling: We construct three different models to predict the target feature "death_yn":

   • Linear Regression
   • Logistic Regression
   • Random Forest

Each model is evaluated based on classification accuracy, confusion matrix, precision, recall, F1 score, and other relevant metrics on both the training set and a hold-out test set. Cross-validation techniques are also employed to assess model robustness.

3. Model Optimization: The final section focuses on improving the predictive models. We explore various strategies such as feature selection, re-scaling, and combining models to enhance accuracy and performance.

File Structure

• Merging and Cleaning: The initial raw CSV data files are merged and cleaned to prepare for analysis.
• Feature Selection: Based on the cleaned dataset, we select relevant features to build our models.
• Model Analysis: Each predictive model is constructed and analyzed for its effectiveness in predicting the target feature.
• Optimization: Efforts to optimize each model are detailed and implemented.