Machine Learning on Covid-19

Machine Learning on Covid-19

Introduction

• In 2020 and the future, covid-19 will be a hot topic and a big challenge for our humans. For my health and life, I am tracking the covid status all the time. For the machine learning practice, I am going to make a project related to Covid-19. In this project, I'll be working with the Covid-19 dataset until 2020/11/10 from https://covidtracking.com/data/download.
• Business problems: The dataset are sorted by every state daily. I am going to look for the dangerous states firstly. Then, I will analyze it deeper in New York State to see how the status of this state now. At last, I will use machine learning to make a classification model predict the covid-19 quality grade daily.

1. The Dangerous States
   • Find some dangerous state by death case rate.
   • Find some dangerous state by positive case rate.
2. The Status now of New York State
   • Visualization of the death cases increasing daily.
   • Visualization of the positive cases increasing daily.
3. Classification model to predict the covid-19 quality grade daily
   • Use logistic regression, random forest, and XGBoost for three different models
   • Use validation set and gridsearchCV for hyper tuning.
   • Use SMOTE for balancing sample sizes

Techniques

• Data collecting, Data cleaning, Exploratory data analysis, machine learning and Visualization are all in the Jupyter Notebook.
• A PowerPoint and pdf for presentation
• A 5 minutes recording to review
• https://youtu.be/D8rI45FBxq4
• Blog: https://road2dataanalyst.blogspot.com/2020/11/machine-learning-on-covid-19.html

Recommendation

• For residents: Even though covid-19 of some states are stable, we still can not relax our guard. Covid-19 is a serious problem in our lives and spreads very quickly. In some dangerous states, such as New York State, there is a risk of rebound. What we can do now is staying at home and keep safe. Do not travel around, especially in some dangerous states.

• For analysts: GridsearchCV is a handy tool to help us tune model parameters, but the shortcoming is costing a long run time.

Future Work

• Accuracy: Find a better or some better parameters to improve accuracy
• Overfitting: Deal with the overfitting of the random forest model and XGBoost model
• More analyses: Try to analyze other columns of the dataset, such as recover

Summary

From this project, I have a new overview of covid-19 and realize how serious of that. Those not just number and data, but human lives. What's more, I had the experience that applying machine learning and logistic regression in our real world. It provided me a chance to review my knowledge and skills. I found my weakness and make a plan to improve those.