数据挖掘（Python）——利用sklearn进行数据挖掘，实现算法：svm、knn、C5.0、NaiveBayes

<span style="line-height: 18px; font-family: Consolas, 'Courier New', Courier, mono, serif; background-color: rgb(255, 255, 255);">说明：下面程序将利用python中的sklearn包根据用户流失前的行为，对用户的流失进行预测</span>

import xlrd
import string
import sklearn
from sklearn import svm
from sklearn import neighbors
from sklearn import cluster
from sklearn import tree
from sklearn import naive_bayes
import numpy as np

#--------数据加载开始--------------
def data_import(filepath):
data=xlrd.open_workbook(filepath)
table=data.sheet_by_index(0)
nrows=table.nrows
ncols=table.ncols
result=[]
result.append(table.row_values(0))
for i in range(1,nrows):
result_0=[]
for j in range(0,ncols):
result_0=result_0+[table.cell(i,j).value]
result.append(result_0)
result=result[1:]
return result

filepath1=r'F:\data_retain.xls'
result1=data_import(filepath1)
count1=len(result1)
target1=[]
data1=[]
for i in range(0,count1):
data1.append(result1[i][1:])
target1.append(result1[i][0])

filepath2=r'F:\data_lost.xls'
result2=data_import(filepath2)
count2=len(result2)
target2=[]
data2=[]
for i in range(0,count2):
data2.append(result2[i][1:])
target2.append(result2[i][0])

data=[]
target=[]
len_1=10000
for i in range(0,len_1):
data.append(data1[i])
target.append(target1[i])

len_2=10000
for i in range(0,len_2):
data.append(data2[i])
target.append(target2[i])
#--------数据加载完毕--------------

class machine_learn:
def svm_(self,data,target):
#支持向量机算法预测
svc = svm.SVC(kernel='linear')
svc.fit(data,target)
predict_target=svc.predict(data)
temp0_0=0
temp0_1=0
temp1_0=0
temp1_1=0
for i in range(0,len(target)):
if predict_target[i]==0:
if target[i]==0:
temp0_0=temp0_0+1
else:
temp0_1=temp0_1+1
else:
if target[i]==0:
temp1_0=temp1_0+1
else:
temp1_1=temp1_1+1
kind_0_precision=temp0_0/(temp0_0+temp0_1)
kind_0_penetration=temp0_0/(temp0_0+temp1_0)
kind_1_precision=temp1_1/(temp1_0+temp1_1)
kind_1_penetration=temp1_1/(temp0_1+temp1_1)
return [[kind_0_precision,kind_0_penetration],[kind_1_precision,kind_1_penetration]]

def knn_(self,data,target):
#最邻近算法算法预测
knn = neighbors.KNeighborsClassifier()
knn.fit(data,target)
predict_target=knn.predict(data)
temp0_0=0
temp0_1=0
temp1_0=0
temp1_1=0
for i in range(0,len(target)):
if predict_target[i]==0:
if target[i]==0:
temp0_0=temp0_0+1
else:
temp0_1=temp0_1+1
else:
if target[i]==0:
temp1_0=temp1_0+1
else:
temp1_1=temp1_1+1
kind_0_precision=temp0_0/(temp0_0+temp0_1)
kind_0_penetration=temp0_0/(temp0_0+temp1_0)
kind_1_precision=temp1_1/(temp1_0+temp1_1)
kind_1_penetration=temp1_1/(temp0_1+temp1_1)
return [[kind_0_precision,kind_0_penetration],[kind_1_precision,kind_1_penetration]]

def tree_(self,data,target):
#决策树算法预测
tre = tree.DecisionTreeClassifier()
tre.fit(data,target)
predict_target=tre.predict(data)
temp0_0=0
temp0_1=0
temp1_0=0
temp1_1=0
for i in range(0,len(target)):
if predict_target[i]==0:
if target[i]==0:
temp0_0=temp0_0+1
else:
temp0_1=temp0_1+1
else:
if target[i]==0:
temp1_0=temp1_0+1
else:
temp1_1=temp1_1+1
kind_0_precision=temp0_0/(temp0_0+temp0_1)
kind_0_penetration=temp0_0/(temp0_0+temp1_0)
kind_1_precision=temp1_1/(temp1_0+temp1_1)
kind_1_penetration=temp1_1/(temp0_1+temp1_1)
return [[kind_0_precision,kind_0_penetration],[kind_1_precision,kind_1_penetration]]

def bayes_(self,data,target):
#使用贝叶斯算法预测（GaussianNB）
bayes = naive_bayes.GaussianNB()
bayes.fit(data,target)
predict_target=bayes.predict(data)
temp0_0=0
temp0_1=0
temp1_0=0
temp1_1=0
for i in range(0,len(target)):
if predict_target[i]==0:
if target[i]==0:
temp0_0=temp0_0+1
else:
temp0_1=temp0_1+1
else:
if target[i]==0:
temp1_0=temp1_0+1
else:
temp1_1=temp1_1+1
kind_0_precision=temp0_0/(temp0_0+temp0_1)
kind_0_penetration=temp0_0/(temp0_0+temp1_0)
kind_1_precision=temp1_1/(temp1_0+temp1_1)
kind_1_penetration=temp1_1/(temp0_1+temp1_1)
return [[kind_0_precision,kind_0_penetration],[kind_1_precision,kind_1_penetration]]

a=machine_learn()
svm_method=a.svm_(data,target)
print ('svm 方法预测，它的准确率可以达到 %f, 覆盖率达到 %f' % (svm_method[1][0],svm_method[1][1]))
knn_method=a.knn_(data,target)
print ('knn 方法预测，它的准确率可以达到 %f, 覆盖率达到 %f' % (knn_method[1][0],knn_method[1][1]))
tree_method=a.tree_(data,target)
print ('决策树 方法预测，它的准确率可以达到 %f, 覆盖率达到 %f' % (tree_method[1][0],tree_method[1][1]))
bayes_method=a.bayes_(data,target)
print ('NaiveBayes 方法预测，它的准确率可以达到 %f, 覆盖率达到 %f' % (bayes_method[1][0],bayes_method[1][1]))

1、当训练样本中，流失用户：留存用户调整为1:1时，决策树模型的前10个规则可以通过覆盖20%的频道用户来定位60%的流失用户

2、<span style="font-family: Arial, Helvetica, sans-serif;">当训练样本中，流失用户：留存用户调整为1:3时，决策树模型的前10个规则可以通过覆盖25.3%的频道用户来定位65.6%的流失用户</span>

<span style="font-family: Arial, Helvetica, sans-serif;"></span><pre name="code" class="python">3、当训练样本中，流失用户：留存用户调整为1:1时，增加部分变量，决策树模型的前10个规则可以通过覆盖15%的频道用户来定位47%的流失用户