RFM analysis with Python

#load data
import pandas as pd
%matplotlib inline
import matplotlib.pyplot as plt
import seaborn as sns
from matplotlib.ticker import FuncFormatter
import datetime as dt
df=pd.read_csv('data.csv',encoding='latin1')

df.head()

df.dtypes

InvoiceNo       object
StockCode       object
Description     object
Quantity         int64
InvoiceDate     object
UnitPrice      float64
CustomerID     float64
Country         object
dtype: object

#transfer data type
df["InvoiceDate"] = pd.to_datetime(df["InvoiceDate"])
df['CustomerID']=df['CustomerID'].astype('object')
df.dtypes

InvoiceNo              object
StockCode              object
Description            object
Quantity                int64
InvoiceDate    datetime64[ns]
UnitPrice             float64
CustomerID             object
Country                object
dtype: object

#create time diff
R_today = dt.datetime(2021,8,7)
df['R_diff'] = (R_today - df['InvoiceDate']).dt.days
df.head(3)

#create R,F,M
R = df.groupby(by=['CustomerID'])['R_diff'].agg([('R','min')])

M=df.groupby(by=['CustomerID'])['UnitPrice'].agg([('M','sum')])

#drop duplicates and calculate the recency
F_undone=df.drop_duplicates(subset='InvoiceNo')
F_undone.head(3)
F=F_undone.groupby(by=['CustomerID'])['CustomerID'].agg([('F','count')])

RFM = R.join(F).join(M)
RFM.head(3)

RFM.describe()

#define R
RFM['R_revalue']=RFM['R']/365
RFM.head(5)

#weight given: R(0.1)/F(0.5)/M(0.4)
RFM['value']=RFM['R_revalue']*0.1+RFM['F']*0.5+RFM['M']*0.4
RFM.head(3)

RFM['value'].describe()
#most of the value under 124
#then we check the distribution of value which under 150

count     4372.000000
mean       132.332697
std        516.430006
min          1.470411
25%         22.965753
50%         54.611945
75%        124.896836
max      16568.499671
Name: value, dtype: float64

RFM_Common=RFM[RFM['value']<150]
sns.set()
sns.distplot(RFM_Common['value'],bins=8)

RFM_Common['value'].std()

38.006808099240864

RFM_Common['value'].mean()

49.523783395743465

#1.96 std contains 95% data, so the top limit should be 136
38+1.96 * 50

136.0

#set four levels of customers, 'level 4' are most valuable 
bins=[0,25,50,136,16569]
RFM['type']=pd.cut(RFM['value'],bins,labels=['level 1','level 2','level 3','level 4'])

RFM['type'].value_counts().plot(kind='pie',autopct='%1.2f%%',subplots=True)

array([],
      dtype=object)

#export customer list
df.to_csv('rfm.csv')