We will use a Kaggle dataset that shows different socio-economic & health factors for different countries. The dataset can be downloaded at this link:...

We will use a Kaggle dataset that shows different socio-economic & health factors for different countries. The dataset can be downloaded at this link:

• 
  https://www.kaggle.com/rohan0301/unsupervised-learning-on-country-data(Links to an external site.)
  

Download the file, open it and check what the different columns represent.

Lab Instructions

KMEANS

Write a python script that clusters the countries using KMeans. Your code shoud:

1. Load the data file into a data frame.


2. Separate the first column (which holds the country name) from the rest of the columns.


3. Run KMeans on the other columns. Set K = 2.


4. Extract the resulting cluster ID's as a list and appending them side by side to the country names. The output from this step should be a dataframe that contains two columns and it should look something like this:
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   

   Country	ClusterID
   Afghanistan	0
   Albania	1
   ...	0

   
   


5. Sort this dataframe by ClusterID & save it to file as a csv file.


6. Open this csv file & see which countries are grouped together into the same cluster.

Do you notice anything interesting about how the different countries are grouped into clusters?

Repeat the clustering process for k = 3, 4 & 6.

For each value of K, show the resulting country clusters, sorted by cluster ID.

Agglomerative

• Repeat the activity above using Agglomerative Clustering. You still need to set the number of clusters through the n_clusters parameter.

DBSCAN

• Run the DBSCAN clustering method with epsilon = 800 and min_samples = 3.


• Check the output file. Notice that when the cluster ID is -1, this means that this is a noise point that is not assigned to any cluster.


• What is the number of obtained clusters?


• How many data points are considered as noise?


• How do you assess the obtained clustering results as compared to the output generated by KMEANS & Agglomerative Clustering?


• Re-run DBSCAN for epsilon values: [700, 800,900] and min_samples values: [2,3]. You can do that by writing two nested for loops of this form:

for eps in range(700,1000,100):
for min_samples in range(2,4,1):
run DBSCAN for eps & min_samples

• Are the obtained results for these values different or similar to each other?

Resources

SciKit Learn Documentation Pages

• KMeans:
  
  
  • 
    https://scikit-learn.org/stable/modules/clustering.html#k-means(Links to an external site.)
    
  
  
  • 
    https://scikit-learn.org/stable/modules/generated/sklearn.cluster.KMeans.html(Links to an external site.)
    
  
  
  
  


• Hierarchical (Agglomerative):
  
  
  • 
    https://scikit-learn.org/stable/modules/clustering.html#hierarchical-clustering(Links to an external site.)
    
  
  
  • 
    https://scikit-learn.org/stable/modules/generated/sklearn.cluster.AgglomerativeClustering.html#sklearn.cluster.AgglomerativeClustering(Links to an external site.)
    
  
  
  
  


• DBSCAN:
  
  
  • 
    https://scikit-learn.org/stable/modules/clustering.html#dbscan(Links to an external site.)
    
  
  
  • 
    https://scikit-learn.org/stable/modules/generated/sklearn.cluster.DBSCAN.html?highlight=clustering%20dbscan(Links to an external site.)