13 RFM 模型

In [413]:

import numpy as np

import pandas as pd

In [414]:

data = pd.read_excel('PYTHON-RFM实战数据.xlsx')

data.head()

Out[414]:

In [416]:

data.info()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 28833 entries, 0 to 28832
Data columns (total 9 columns):
品牌名称    28833 non-null object
买家昵称    28833 non-null object
付款日期    28833 non-null datetime64[ns]
订单状态    28833 non-null object
实付金额    28833 non-null int64
邮费      28833 non-null int64
省份      28833 non-null object
城市      28832 non-null object
购买数量    28833 non-null int64
dtypes: datetime64[ns](1), int64(3), object(5)
memory usage: 2.0+ MB

In [417]:

#仅当交易成功时 ，才有FRM模型

data = data[data.订单状态=='交易成功']

data.shape

Out[417]:

(27793, 9)

13.1 关键字段提取

In [418]:

data = data[['买家昵称','付款日期','实付金额','购买数量']]

data.head()

Out[418]:

13.1.1 R

In [419]:

# 当前观测时间（数据提取时间）假设 是2019-07-1

data.付款日期.max()

Out[419]:

Timestamp('2019-06-30 22:46:22.511000')

In [420]:

data.买家昵称.shape

Out[420]:

(27793,)

In [422]:

data.买家昵称.nunique()

Out[422]:

25420

In [423]:

#由于同一个客户对应不同的订单，选取客户付款日期的最大值

r = data.groupby('买家昵称')['付款日期'].max().reset_index()

r.head()

Out[423]:

In [424]:

# 时间日期格式可以直接做差值

r['R'] = (pd.to_datetime('2019-7-1') - r.付款日期).dt.days

r

Out[424]:

25420 rows × 3 columns

In [425]:

# 删除无用特征

r = r[['买家昵称','R']]

r.head()

Out[425]:

13.1.2 F

In [426]:

f = data.groupby('买家昵称')['付款日期'].count().reset_index()

f.columns=['买家昵称','F']

f.head()

Out[426]:

13.1.3 M

In [428]:

# 总金额

m = data.groupby('买家昵称')['实付金额'].sum().reset_index()

m.columns=['买家昵称','M']

m.head()

Out[428]:

13.2 三张表合并

In [431]:

rfm = pd.merge(r,f)

rfm = pd.merge(rfm,m)

rfm

Out[431]:

25420 rows × 4 columns

13.3 维度打分

In [432]:

rfm.describe().T

Out[432]:

In [433]:

# 采用5分制对rfm进行打分

# R 0-30-60-90-120-以上

rfm['R-score'] = pd.cut(rfm['R'],bins=[-1,30,60,90,120,180],labels=[5,4,3,2,1]).astype(float)

rfm.head()

Out[433]:

In [436]:

# F

rfm['F-score'] = pd.cut(rfm['F'],bins=[1,2,3,4,5,20],labels=[1,2,3,4,5],right=False).astype(float)

# M

rfm['M-score'] = pd.cut(rfm['M'],bins=[0,50,100,150,200,10000],labels=[1,2,3,4,5],right=False).astype(float)

In [437]:

rfm.head()

Out[437]:

In [434]:

rfm.F.value_counts()

Out[434]:

1     23367
2      1795
3       217
4        33
5         5
15        1
7         1
6         1
Name: F, dtype: int64

In [435]:

rfm.M.value_counts()

Out[435]:

87     1289
94      746
144     658
34      643
89      634
       ... 
451       1
483       1
515       1
531       1
687       1
Name: M, Length: 536, dtype: int64

13.3.1 第二轮打分：和第一轮分数平均值作比较

In [438]:

rfm['R是否大于均值'] = (rfm['R-score'] > rfm['R-score'].mean())*1

rfm['F是否大于均值'] = (rfm['F-score'] > rfm['F-score'].mean())*1

rfm['M是否大于均值'] = (rfm['M-score'] > rfm['M-score'].mean())*1

In [439]:

rfm

Out[439]:

25420 rows × 10 columns

13.4 客户分层

In [444]:

# 通过数值的加法rfm组合在一起

rfm['人群数值'] = rfm['R是否大于均值']*100+rfm['F是否大于均值']*10+rfm['M是否大于均值']

rfm.head()

Out[444]:

In [443]:

rfm.head()

Out[443]:

13.5 基于人群的数值为客户打上标签

In [445]:

def trans(x):

    if x==111:

        label = '高价值客户'

    elif x==11:

        label = '重要召回客户'

    elif x==101:

        label = '重要发展客户'

    elif x==1:

        label = '重要挽留客户'

    elif x==110:

        label = '潜力客户'

    elif x==100:

        label = '新客户'

    elif x==10:

        label = '一般客户'

    elif x==0:

        label = '流失客户'

    return label

In [446]:

rfm['人群类型'] = rfm['人群数值'].apply(trans)

rfm

Out[446]:

25420 rows × 12 columns

13.5.1 统计描述：人群统计、金额统计

In [447]:

count = rfm['人群类型'].value_counts().reset_index()

count.columns=['客户类型','人数']

count

Out[447]:

In [448]:

rfm['M']

Out[448]:

0         49
1        183
2         34
3         37
4        164
        ... 
25415    175
25416     87
25417    497
25418    137
25419    185
Name: M, Length: 25420, dtype: int64

In [449]:

money = rfm.groupby('人群类型')['M'].sum().reset_index()

In [450]:

money

Out[450]: