Note: There is an explanation of the project and the results below.
Today, the world is facing a widespread spread of the corona virus and an increase in the number of infections and deaths due to this disease, wishing safety and health for all.
Note: This project is a research study and we are working on developing it and increasing the number of features used in training, as we only used today's number in the training process
Aim : Predict using ML the number of infected people and the number of deaths of coronavirus.
Run:
-
You can make clone for this repo and then run via Jupyter.
Another way is faster:
- Go to the project's Colaboratory from here
- Make your own copy for run.(File --> Save a copy in drive )
- Run cell
Technologies:
DataSet : We used data for the United States to train our model and we got data for 95 days that includes the date of the day and the number of people with the disease and for the data of the dead we got data for 60 days and also includes the date of the day and the number of deaths due to disease.
The reference worldometers
We used 3 types of ML models :
- Linear regression with log values
- Support Vector Regression (SVR)
- Artificial Neutral Network (MLPRegressor)
- The data was divided into two sections, for training and testing .
- We adopted the K-Fold Cross-validation methodology when selecting
the best parameter for a single model.
The following figure shows the work methodology:
When looking for data for the United States - Active Cases (Total), we obtained data for 95 days, starting 21/1/2020 , and the number of injured was one and we assumed on the first day. The data ended on 25/4/2020, today is 95, and the number of injured is 788,233.
We took data for 12 days randomly for the final test and evaluation stage. After that we trained the different models, got a final evaluation and represented them with score for each model.
The following figure represents the curve for each model in addition to the score obtained:
Note: The data were represented using two methods, the first is a logarithmic representation and the second is a Linear representation (Exponential). The models were trained in both methods, and we chose the best among them for each model individually.
- When looking at the Linear Regression curve, we used logarithmic representation of data when training it, and the resulting curve was converted to the exponential representation.
Linear representation (Exponential)
logarithmic representation
After looking at the results, we realize that the best model is the Artificial Neutral Network (MLPRegressor), and it has the best score !
When searching for data for the United States, we obtained data for 60 days, starting on 2/3/2020, and the number of dead was 6 and we assumed on the first day. The data ended on 30/4/2020, today 60, and the number of dead 63856.
We took data for 8 days randomly for the final test and evaluation stage. After that we trained the different models, got a final evaluation and represented them with score for each model.
The following figure represents the curve for each model in addition to the score obtained:
Note: The data were represented using two methods, the first is a logarithmic representation and the second is a Linear representation (Exponential). The models were trained in both methods, and we chose the best among them for each model individually.
- When looking at the Linear Regression curve, we used logarithmic representation of data when training it, and the resulting curve was converted to the exponential representation.
Linear representation (Exponential)
logarithmic representation
After looking at the results, we realize that the best model is the Artificial Neutral Network (MLPRegressor), and it has the best score !