I am working on the AdventureWorks database and I have extracted some demographic data from the person scheme as follow. My aim is to create a customer segmentation with the demographic data using the clustering technique.
| businessentityid | YearlyIncome | Education | Occupation | HomeOwnerFlag |
|---|---|---|---|---|
| 1699 | 25001-50000 | Graduate Degree | Clerical | 1 |
| 1700 | 50001-75000 | Bachelors | Professional | 0 |
Because the dataframe contains categorical data, I have decided to encode it using OneHotEncoder. The resulting table is as follows. Note that the tables and dataframes provided is just a sample in order to illustrate the idea.
encoder = OneHotEncoder(handle_unknown='ignore')
| YearlyIncome_0-25000 | YearlyIncome_25001-50000 | Education_Graduate Degree | Education_Bachelors | HomeOwnerFlag_0 | HomeOwnerFlag_1 |
|---|---|---|---|---|---|
| 0 | 1 | 0 | 0 | 1 | 0 |
| 1 | 0 | 0 | 1 | 0 | 1 |
Later down the line, I used Principal Component Analysis to reduce the dimensions.
pca = PCA(n_components=2)
And then applied KMeans algorithm. I experimented with elbow method to find out that the 4 is optimal n value.
kmeans = KMeans(n_clusters=4)
print(kmeans4.labels_)
array([0, 1, 3, ..., 1, 0, 2], dtype=int32)
Lastly, I combined the initial dataframe with the labels.
| businessentityid | Label |
|---|---|
| 1699 | 0 |
| 1700 | 1 |
| 1701 | 3 |
How do I interpret the labels? There are four clusters but in the process of encoding with OneHotEncoder, the dimensions have increased and got complicated. The PCA reduced the dimensions to 2 (could be different), but how do I know what label means what? Is there something else to follow?
