SQL项目实战：银行客户分析

大家好，本文将与大家分享一个SQL项目，即根据从数据集收集到的信息分析银行客户流失的可能性。这些洞察来自个人信息，如年龄、性别、收入和人口统计信息、银行卡类型、产品、客户信用评分以及客户在银行的服务时间长短等。对于银行而言，了解如何留住客户比寻找其他客户更有利。

客户流失是指客户或顾客的流失。公司通常将其作为关键业务指标之一，因为恢复的长期客户对公司的价值远远高于新招募的客户。客户流失有两种类型：自愿流失和非自愿流失。自愿流失是由于客户决定转向其他公司或服务提供商，而非自愿流失则是由于客户搬迁到长期护理机构、死亡或搬迁到较远的地方等情况造成的。

本文将集中讨论自愿流失，因为它可能是由于公司与客户关系中公司可以控制的因素造成的，例如如何处理账单互动或如何提供售后帮助。

数据集

本文使用的是customer_churn_records表，该表包含多列，customerid是表的主键。

RowNumber：对应记录（行）编号
CustomerId：客户的ID编号
Surname：客户的姓氏
CreditScore：客户信用行为预测值
Geography：客户所在地
Gender：客户的性别信息
Age：客户的年龄信息
Tenure：客户在银行的使用年限
Balance：客户账户中的余额信息
NumOfProducts：客户购买的产品数量
HasCrCard：客户是否拥有信用卡
IsActiveMember：客户是否处于活跃状态
EstimatedSalary（估计工资）：客户的估计工资金额
Exited：客户是否离开银行
Complain：客户是否有投诉
Satisfaction Score：客户对银行的满意度评分
Card Type：客户持有的银行卡类型
Points Earned：客户使用信用卡获得的积分

# 显示表中的列 = customer_churn_recordsq='''SELECT * FROM customer_churn_records'''df = pd.read_sql(q,engine_postgresql)
df.head()

查询客户流失率

导入软件包：

import psycopg2 # PostgreSQL数据库适配器
import pandas as pd # 用于分析数据
from sqlalchemy import create_engine # 促进Python程序与数据库之间的通信

首先，本文根据已退出的列计算有多少客户流失。

# 统计是否流失/退出的客户总数q='''WITH temp_churn AS(SELECT exited,CASE WHEN exited = 1 THEN 'Churn'ELSE 'Not Churn'END AS STATUSfrom customer_churn_records)SELECT STATUS,COUNT(exited) as TotalFROM temp_churnGROUP BY 1'''df = pd.read_sql(q,engine_postgresql)
df.head()

可以发现在10000名客户中，有近20%从银行退出或流失。尽管这个数字并不算很大，但如果现在还不能解决这个问题，它可能会增长得更多。

现在，本文将从活跃客户、性别、人口统计、年龄、临时工龄、信用分数、产品数量、满意度分数、投诉、是否有信用卡、卡类型、已获积分、预估薪资和余额等多个方面来检查客户流失状况的类型。

# 统计有多少活跃客户流失q=''' WITH temp_isactivemember AS(SELECT exited,CASE WHEN isactivemember = 1 THEN 'Active'ELSE 'Not Active'END AS isactivememberfrom customer_churn_records)SELECT isactivemember,COUNT (CASE WHEN exited = 1 THEN 1 END) AS Churn,COUNT (CASE WHEN exited = 0 THEN 1 END) AS Not_ChurnFROM temp_isactivememberGROUP BY 1'''df = pd.read_sql(q,engine_postgresql)
df

# 根据性别计算是否流失/退出的客户总数q='''SELECT gender,COUNT(gender) as Total,COUNT(case when exited = 1 then 1 end) as Churn,COUNT(case when exited = 0 then 1 end) as Not_churnFROM customer_churn_recordsGROUP BY 1'''df = pd.read_sql(q,engine_postgresql)
df.head()

# 根据人口统计数据计算流失客户的数量q=''' SELECT geography,COUNT (CASE WHEN exited = 1 THEN 1 END) AS Churn,COUNT (CASE WHEN exited = 0 THEN 1 END) AS Not_ChurnFROM customer_churn_recordsGROUP BY 1'''df = pd.read_sql(q,engine_postgresql)
df

#根据年龄组计算流失客户的数量q=''' SELECT CASE WHEN age <= 20 THEN 'Group <= 20'WHEN age >= 21 AND age <= 40 THEN 'Group 21-40'WHEN age >= 41 AND age <= 60 THEN 'Group 41-60'ELSE 'Group > 60'END AS age_category,COUNT(CASE WHEN exited = 1 then 1 end) as Churn,COUNT(CASE WHEN exited = 0 then 1 end) as Not_ChurnFROM customer_churn_recordsGROUP BY 1ORDER BY 1'''df = pd.read_sql(q,engine_postgresql)
df

# 根据他们成为客户的时间计算是否流失/退出的客户总数q='''WITH temp_tenure AS(SELECT tenure,CASE WHEN exited = 1 THEN 'Churn'ELSE 'Not Churn'END AS STATUSfrom customer_churn_records)SELECT STATUS,AVG(tenure) as Average_tenureFROM temp_tenureGROUP BY 1'''df = pd.read_sql(q,engine_postgresql)
df.head()

# 根据客户使用银行产品的数量，计算有多少客户流失q='''WITH temp_bankprod AS(SELECT numofproducts,CASE WHEN exited = 1 THEN 'Churn'ELSE 'Not Churn'END AS STATUSfrom customer_churn_records)SELECT STATUS,AVG(numofproducts) as avg_numofproductsFROM temp_bankprodGROUP BY 1'''df = pd.read_sql(q,engine_postgresql)
df.head()

# 根据客户 对银行的满意度得分的平均得分，计算有多少客户流失q='''WITH temp_satisfaction AS(SELECT satisfaction_score,CASE WHEN exited = 1 THEN 'Churn'ELSE 'Not Churn'END AS STATUSfrom customer_churn_records)SELECT STATUS,AVG(satisfaction_score) as satisfaction_levelFROM temp_satisfactionGROUP BY 1'''df = pd.read_sql(q,engine_postgresql)
df.head()

# 根据客户的投诉量，统计有多少客户流失q='''WITH temp_complain AS(SELECT complain,CASE WHEN exited = 1 THEN 'Churn'ELSE 'Not Churn'END AS STATUSfrom customer_churn_records)SELECT STATUS,COUNT(complain) as complainFROM temp_complainGROUP BY 1'''df = pd.read_sql(q,engine_postgresql)
df.head()

# 根据客户是否拥有信用卡计算有多少客户流失q=''' WITH temp_hascrcard AS(SELECT hascrcard,CASE WHEN exited = 1 THEN 'Churn'ELSE 'Not Churn'END AS STATUSfrom customer_churn_records)SELECT STATUS,COUNT (CASE WHEN hascrcard = 1 THEN 1 END) AS has_creditcard,COUNT (CASE WHEN hascrcard = 0 THEN 1 END) AS no_creditcardFROM temp_hascrcardGROUP BY 1'''df = pd.read_sql(q,engine_postgresql)
df

# 根据客户的预估工资计算有多少客户流失q='''WITH temp_salary AS(SELECT estimatedsalary,CASE WHEN exited = 1 THEN 'Churn'ELSE 'Not Churn'END AS STATUSfrom customer_churn_records)SELECT STATUS,AVG(estimatedsalary) as avg_salary,MAX(estimatedsalary) as max_salary,MIN(estimatedsalary) as min_salaryFROM temp_salaryGROUP BY 1'''df = pd.read_sql(q,engine_postgresql)
df.head()

# 根据客户的银行存款余额计算有多少客户流失
#（平均、最高、最低）q='''WITH temp_balance AS(SELECT balance,CASE WHEN exited = 1 THEN 'Churn'ELSE 'Not Churn'END AS STATUSfrom customer_churn_records)SELECT STATUS,AVG(balance) as avg_balance,MAX(balance) as max_balance,MIN(balance) as min_balanceFROM temp_balanceGROUP BY 1'''df = pd.read_sql(q,engine_postgresql)
df.head()