Inspiration
It has been noticed that many people are afraid that there might be a second outbreak of COVID-19 due to sharp lockdowns in certain districts of some major cities in China. Hence, we were inspired to make a five-day prediction of the case number so that people can have an overview of the spread of COVID-19 and better prepare themselves to cope with it.
What it does
Our project is aimed to predict the daily confirmed/recovery/fatal cases in every country in the following five days, and it is presented in the form of a table and a world map on a website.
How we built it
There are basically three stages in the process of building our project.
Stage I: Brainstorming
We ran a quick brainstorm to determine which track and hot issue we would like to choose by analyzing and discussing all three tracks in different settings (i.e. floods, 2020 Tokyo Olympics, COVID-19, etc). Our final decision is Track I: Prediction in the context of COVID-19.
Stage II: Writing proposals
We came up with three proposals regarding what kind of predictions we would like to make for our target users and relevant applications of those predictions. In this stage, we compared and contrasted strategies (i.e. the lockdown/curfew policy) that major countries have used to cope with COVID-19, and we conducted a thorough market analysis of the economic and social impact of COVID-19. Though there were another two potential proposals available, we eventually chose to predict the daily case number on a worldwide level after taking feasibility and other factors into account.
Stage III: Coding
First, we retrieved data including confirmed/fatal/recovery cases in every country from covid19dh 2.3.0 API and saved results on Excel. Second, we processed and analyzed data by using six tree-based machine learning algorithms - Random Forests, Gradient Boosting, K-Nearest Neighbor Regression (kNN), least absolute shrinkage, and selection operator (LASSO), Ridge, Elastic Net. We also introduced Root Mean Square Error (RMSE) to help us to understand the performance of each algorithm. Then we separated data into a confirmed cases table and a death case table and visualized them by creating a map. Third, we programmed the HTML file to design the basic layout of the website with the code sent from the backend, including user navigation and interface, and analyzed possible requests from users and corresponding functions to respond.
Challenges we ran into
At first, we encountered a few challenges in Stage II: Writing proposals. Though we decided to predict the daily case number in the end, our original idea was to predict the possibility and impact of a lockdown. However, it was tricky to select our target users (those who will benefit most from our predictions - ordinary people V.S. small private business owners) and to access abundant, authenticated data, making it difficult to determine whether we should explore this on a national or international scale. We also found it unrealistic to retrieve and process required data and build a sophisticated algorithmic model within a limited time, as there are several factors needed to be considered when it comes to the possibility and impact of a lockdown.
Then there were also some challenges in Stage III: Coding. Required data for four countries was unavailable, so we neglected those countries. It was difficult to make predictions based on previous predictions, in which unreasonable results might be generated. We also encountered some problems when creating the transition world map as we used different types of data. There were some bugs in our coding program when predicting confirmed/fatal/recovery cases on the fifth day, but we debugged successfully later. It was also difficult to find a suitable API that includes the death number to retrieve COVID-19 data. We also found it confusing to send data from backend to frontend using different coding languages and platforms.
Accomplishments that we're proud of
First, we are proud of our highly efficient teamwork. We brainstormed and determined the theme of our project real quick, allocated tasks based on everyone’s strength, set small deadlines to check the progress of individual tasks, gave constructive feedback, and offered help to each other. Though all of us are from different time zones (AEST, EST, CST, and PST), we still manage to meet up online frequently to update our progress and solve problems together.
Second, we are proud of the completion of our project. Though none of us has prior hackathon experience, we complete our project in a timely, productive manner. Not only have we gained valuable experience in coding and market analysis, but we also feel motivated when realizing our project may make a difference in people’s daily lives.
What we learned
With regards to hard skills, we have learned how to communicate with different coding languages and to send live data from the back-end to the front-end.
Speaking of soft skills, we have learned how to collaborate with people with different needs online. We understood, compromised, and adapted to each other’s needs.
What's next for (COVID-19 Case Number Predictions)
As we mentioned above, we could improve our project by predicting the possibility and impact of city/statewide and nationwide lockdowns based on the prediction of the case number and other relevant factors. We believed that it could be useful for people to minimize their losses by preparing for the potential lockdown in advance.
One idea is to predict the possibility of a city/statewide lockdown and its impact on access to grocery items. Our website will categorize the grocery items based on their usage (such as food, sanitation, appliances), predict their availability, and further make suggestions of a reasonable purchase amount based on the geographical range and duration of the lockdown. In doing so, we can prevent people from necessity shortage at home and purchasing excessive items, causing resource shortage. We would select a country as our target where its residents’ life quality and resource distributions are affected the most due to the lockdown policy since we hope to benefit as many people as possible. We also need additional data such as the number of daily new cases, city populations, the standard for grocery item storage, and scenarios when a lockdown is likely to be announced.
The other idea is to predict the possibility of a nationwide lockdown and its impact on international transportation and traveling in terms of the potential delay in delivery time and price changes, and we will especially make suggestions about air/sea shipping and booking of flights.
For international transportation, we can predict how long the delay would be and how much the price would increase if the goods are transported by air or sea and in turn suggest the need to increase or decrease the need to transport. Thus, we can help users, such as ordinary people who purchase things overseas online and businesses that largely rely on supply-chain (i.e. import/export companies, inventory business, etc), foresee the impact of the disruption of the global supply chain so that they can deal with it at an earlier stage. We need data such as variations of delivery fees since the lockdown, daily delivery fees, variations of delivery time since the lockdown, and daily delivery time.
For international traveling (especially international aviation), we can predict how large the possibility is that the flight might cancel and how much the price would increase and in turn suggest the best timing to book the flight, which can help ordinary people better prepare for international traveling, as flights may be canceled or their prices may increase greatly during the lockdown. We need data such as price variations since the lockdown, current prices, and situations when a cancellation happened due to the lockdown.
Log in or sign up for Devpost to join the conversation.