Covid Tracing using DBSCAN

People who come into contact with someone infected with Covid-19 within a radius of 2m have a high probability of being positive. These people are called F1 (who has been infected is F0).

With the grouping of people in close contact, the DBSCAN algorithm and the Haversine distance formula (used to calculate the distance of 2 points on the sphere) is a pretty suitable choice for this.

Data preparation

There are two datasets required:

• The dataset of coordinates of people in Vietnam.
• The dataset of geometry of the province in Vietnam.

Coordinates dataset

The coordinate will be randomly generated in the bound of Vietnam. This has been done before, so what I need to do now is read the saved CSV file

Geometry dataset

Geojson file of Vietnam can be found in this link

Libraries

Here are the required libraries to be able to run the file:

• numpy
• pandas
• geopandas
• pycristoforo for getting Vietnam bounding box
• sklearn
• plotly
• shapely

Coding process

Data processing

Process coordinates dataset

If there is already a saved CSV file, read it. Otherwise, generate a new one.

Then rename column lo to longitude and la to latitude

df = df.rename(columns={'lo': 'longitude', 'la': 'latitude'})

Assign id to each row as index

df['id'] = df.index

For the column 'infected' it is randomly generated with a rate of 1% positive and 99% negative out of 10,000,000 people

df['infected'] = np.random.choice([True, False],
                                  size=10000000,
                                  p=[0.01, 0.99])

Column F1 initially has the value False

df['f1'] = False

Process geometry dataset

Geometry dataset is a geojson file. Use geopandas to read it.

df_geo = gpd.read_file('./data/diaphantinh.geojson')

DBSCAN

The value to train X is the latitude and longitude that has been converted to radians

X = df[['longitude', 'latitude']].to_numpy()
X = np.radians(X)

Initiate the DBSCAN algorithm with:

• Metric: Haversine distance
• Epsilon:
  • The distance to determine close contact is 2 meters
  • Earth's radius is 6371 km
  • So the epsilon is 2/6371 = 0.000011
• Min_samples: As long as 2 people come into contact with each other, it will be considered a cluster
• Algorithm: ball_tree

dbscan = DBSCAN(metric='haversine',
                eps=0.02/6371,
                min_samples=2,
                algorithm='ball_tree')

Result processing

Create a new DataFrame consisting of the original DataFrame combined with the resulting clusters column after fitting the DBSCAN algorithm

df_result = df.assign(cluster=clusters)

In this DataFrame, this result only needs people who are F0 and those who are close to each other

df_result = df_result[(df_result['cluster'] != -1) |
                      (df_result['infected'] == True)]

People who close contact with someone infected with Covid-19 will be labeled as F1

for index, row in df_result[(df_result['infected'] == True)
                            & (df_result['cluster'] >= 0)].iterrows():
    same_id = retrieve_same_cluster(df_result,
                                    row.id,
                                    exclusive=True).id.values
    df_result.loc[same_id, 'f1'] = True

Result visualization

Close contact visualization

Draw a map of people in close contact with each other

Demo a cluster of people in close contact with each other

Infected visualization

Draw a map according to the number of people infected with Covid in each province

Besides, there is a column chart showing the number of infected people by province sorted from highest to lowest

F1 visualization

Similar to above, draw 2 graphs to show people with F1

Conclusion

Aggregate number of infected people and contacts of infected people represented by grouped bar chart