RFM Clustering example

rfm_cluster.py

import pandas as pd
import numpy as np
from sklearn.cluster import KMeans

df = pd.read_csv('user_history.csv')

# Pandas' cut method groups continuous values into equal-sized bins
df['Frequency'] = pd.cut(df['RequestsPerMonth'], bins=4)

# Since lower recency is better, we need to reverse the order of the bins
df['Recency'] = pd.cut(df['DaysSinceLastRequest'], bins=4)
df['Recency'] = df['Recency'].replace({
    3: 0,
    2: 1,
    1: 2,
    0: 3
})
df['MonetaryValue'] = pd.cut(df['OrderTotal'], bins=4)

# Calculate value score based on these metrics:
df['ValueScore'] = df[['Frequency', 'Recency', 'MonetaryValue']].sum(axis=1)

# Cluster users with K-means based on their value score by High, Medium, and Low value
# Indices of clusters are random, so we'll need to order them by value
model = KMeans(n_clusters=3)
df['ValueCluster'] = model.fit_predict(df[['ValueScore']])

# Get the 1-D centroid of each cluster
centroids = df.groupby('ValueCluster')['ValueScore'].mean().reset_index(drop=True)

# Mapping should be in order of value
cluster_mapping = np.argsort(centroids)

# Replace random cluster indices with value-ordered indices
df['ValueCluster'] = df['ValueCluster'] \
    .replace({val: key for key, val in enumerate(cluster_mapping)})