电信用户建模
数据来源
电信用户数据：https://www.datafountain.cn/dataSets/35/details#
1.数据概述与可视化
# 模块 import pandas as pd import numpy as np import matplotlib.pyplot as plt import seaborn as sns import scipy.stats as st plt.rc("font",family="SimHei",size="12") #解决中文无法显示的问题 # 导入数据,且要赋值一份数据，用于数据可视化 train=pd.read_csv('F:\\python\\电信用户数据\\WA_Fn-UseC_-Telco-Customer-Churn.csv')
train_data=train.copy()
1.1 数据概述
train_data.head() train_data.describe() train_data.info() train_data.shape #(7043, 21)
<class 'pandas.core.frame.DataFrame'>
RangeIndex: 7043 entries, 0 to 7042
Data columns (total 21 columns):
# Column Non-Null Count Dtype
--- ------ -------------- -----
0 customerID 7043 non-null object
1 gender 7043 non-null object
2 SeniorCitizen 7043 non-null int64
3 Partner 7043 non-null object
4 Dependents 7043 non-null object
5 tenure 7043 non-null int64
6 PhoneService 7043 non-null object
7 MultipleLines 7043 non-null object
8 InternetService 7043 non-null object
9 OnlineSecurity 7043 non-null object
10 OnlineBackup 7043 non-null object
11 DeviceProtection 7043 non-null object
12 TechSupport 7043 non-null object
13 StreamingTV 7043 non-null object
14 StreamingMovies 7043 non-null object
15 Contract 7043 non-null object
16 PaperlessBilling 7043 non-null object
17 PaymentMethod 7043 non-null object
18 MonthlyCharges 7043 non-null float64
19 TotalCharges 7043 non-null object
20 Churn 7043 non-null object
dtypes: float64(1), int64(2), object(18)
memory usage: 1.1+ MB
| SeniorCitizen | tenure | MonthlyCharges | |
|---|---|---|---|
| count | 7043.000000 | 7043.000000 | 7043.000000 |
| mean | 0.162147 | 32.371149 | 64.761692 |
| std | 0.368612 | 24.559481 | 30.090047 |
| min | 0.000000 | 0.000000 | 18.250000 |
| 25% | 0.000000 | 9.000000 | 35.500000 |
| 50% | 0.000000 | 29.000000 | 70.350000 |
| 75% | 0.000000 | 55.000000 | 89.850000 |
| max | 1.000000 | 72.000000 | 118.750000 |
由此可见很多数据都是对象型的，数值型很少，但是我们也要注意,居然没有缺失值，但是我们也要注意非空值的缺失值，类似于'-'这种的
1.2 数据可视化
首先区分一下数值型特征和类别型特征
#类别特征 object_col=dianxin.select_dtypes(include=[np.object]).columns ''' Index(['customerID', 'gender', 'Partner', 'Dependents', 'PhoneService', 'MultipleLines', 'InternetService', 'OnlineSecurity', 'OnlineBackup', 'DeviceProtection', 'TechSupport', 'StreamingTV', 'StreamingMovies', 'Contract', 'PaperlessBilling', 'PaymentMethod', 'TotalCharges', 'Churn'], dtype='object') ''' #数值特征 num_col=dianxin.select_dtypes(include=[np.number]).columns ''' Index(['SeniorCitizen', 'tenure', 'MonthlyCharges'], dtype='object') '''
1.2.1单变量可视化（类别特征）
#单变量可视化 #首先看一些大体的分布如何，后面才好画图 for i in object_col: print(i,train_data[i].nunique())
customerID 7043 gender 2 Partner 2 Dependents 2 PhoneService 2 MultipleLines 3 InternetService 3 OnlineSecurity 3 OnlineBackup 3 DeviceProtection 3 TechSupport 3 StreamingTV 3 StreamingMovies 3 Contract 3 PaperlessBilling 2 PaymentMethod 4 TotalCharges 6531 Churn 2
#customerID，TotalCharges,可以不用画图 object_col=['gender', 'Partner', 'Dependents', 'PhoneService', 'MultipleLines', 'InternetService', 'OnlineSecurity', 'OnlineBackup', 'DeviceProtection', 'TechSupport', 'StreamingTV', 'StreamingMovies', 'Contract', 'PaperlessBilling', 'PaymentMethod', 'Churn'] fig = plt.figure(figsize=[16,40]) for col,i in zip(object_col,range(1,17)): axes = fig.add_subplot(9,2,i) sns.countplot(train_data[col],ax=axes) plt.show()
有上图可知，除了PhoneService分布特别不合理外，其余的分布还是比较均衡的
但是我们注意到下面，有三个值（类似于No phone service和No）其实是二个值，'MultipleLines' , 'OnlineSecurity', 'OnlineBackup', 'DeviceProtection', 'TechSupport', 'StreamingTV', 'StreamingMovies'
for i in object_col: for j in train_data[i].value_counts(normalize = True).index: print(i,'=',j,":",round(train_data[i].value_counts(normalize = True)[j]*100,2),'%') print(' ')
gender = Male : 50.48 %
gender = Female : 49.52 %
Partner = No : 51.7 %
Partner = Yes : 48.3 %
Dependents = No : 70.04 %
Dependents = Yes : 29.96 %
PhoneService = Yes : 90.32 %
PhoneService = No : 9.68 %
MultipleLines = No : 48.13 %
MultipleLines = Yes : 42.18 %
MultipleLines = No phone service : 9.68 %
InternetService = Fiber optic : 43.96 %
InternetService = DSL : 34.37 %
InternetService = No : 21.67 %
OnlineSecurity = No : 49.67 %
OnlineSecurity = Yes : 28.67 %
OnlineSecurity = No internet service : 21.67 %
OnlineBackup = No : 43.84 %
OnlineBackup = Yes : 34.49 %
OnlineBackup = No internet service : 21.67 %
DeviceProtection = No : 43.94 %
DeviceProtection = Yes : 34.39 %
DeviceProtection = No internet service : 21.67 %
TechSupport = No : 49.31 %
TechSupport = Yes : 29.02 %
TechSupport = No internet service : 21.67 %
StreamingTV = No : 39.9 %
StreamingTV = Yes : 38.44 %
StreamingTV = No internet service : 21.67 %
StreamingMovies = No : 39.54 %
StreamingMovies = Yes : 38.79 %
StreamingMovies = No internet service : 21.67 %
Contract = Month-to-month : 55.02 %
Contract = Two year : 24.07 %
Contract = One year : 20.91 %
PaperlessBilling = Yes : 59.22 %
PaperlessBilling = No : 40.78 %
PaymentMethod = Electronic check : 33.58 %
PaymentMethod = Mailed check : 22.89 %
PaymentMethod = Bank transfer (automatic) : 21.92 %
PaymentMethod = Credit card (automatic) : 21.61 %
Churn = No : 73.46 %
Churn = Yes : 26.54 %
#但是我们注意到下面，有三个值（类似于No phone service和No）其实是二个值 two_no_col=['MultipleLines' , 'OnlineSecurity', 'OnlineBackup', 'DeviceProtection', 'TechSupport', 'StreamingTV', 'StreamingMovies'] dict1={'No phone service':'No', 'No':'No', 'Yes':'Yes'} train_data['MultipleLines']=train_data['MultipleLines'].map(dict1) dict2={'No internet service':'No', 'No':'No', 'Yes':'Yes'} for i in ['OnlineSecurity', 'OnlineBackup', 'DeviceProtection', 'TechSupport', 'StreamingTV', 'StreamingMovies']: train_data[i]=train_data[i].map(dict2) fig = plt.figure(figsize=[16,20]) for col,i in zip(two_no_col,range(1,8)): axes = fig.add_subplot(4,2,i) sns.countplot(train_data[col],ax=axes) plt.show() for i in two_no_col: for j in train_data[i].value_counts(normalize = True).index: print(i,'=',j,":",round(train_data[i].value_counts(normalize = True)[j]*100,2),'%') print(' ')
MultipleLines = No : 57.82 %
MultipleLines = Yes : 42.18 %
OnlineSecurity = No : 71.33 %
OnlineSecurity = Yes : 28.67 %
OnlineBackup = No : 65.51 %
OnlineBackup = Yes : 34.49 %
DeviceProtection = No : 65.61 %
DeviceProtection = Yes : 34.39 %
TechSupport = No : 70.98 %
TechSupport = Yes : 29.02 %
StreamingTV = No : 61.56 %
StreamingTV = Yes : 38.44 %
StreamingMovies = No : 61.21 %
StreamingMovies = Yes : 38.79 %
1.2.2单变量和y值可视化（类别特征）
但是要注意，barplot需要y值是数值型的，现在我们先把y值转化成数值型的，Churn=1即是逾期
#类别变量和y值的可视化 #但是要注意，barplot需要y值是数值型的，现在我们先把y值转化成数值型的，Churn=0即是逾期 train_data.Churn[train_data['Churn']=='No']=0 train_data.Churn[train_data['Churn']=='Yes']=1 fig = plt.figure(figsize=[16,40]) for col,i in zip(object_col,range(1,17)): axes = fig.add_subplot(8,2,i) sns.barplot(train_data[col],train_data['Churn'],ax=axes) plt.show()
#不要y值 col=['gender', 'Partner', 'Dependents', 'PhoneService', 'MultipleLines', 'InternetService', 'OnlineSecurity', 'OnlineBackup', 'DeviceProtection', 'TechSupport', 'StreamingTV', 'StreamingMovies', 'Contract', 'PaperlessBilling', 'PaymentMethod',] for i in col: for j in train_data[i].value_counts(normalize = True).index: print(i,'=',j,"的违约概率:",round(train_data.Churn[train_data[i]==j].value_counts(normalize = True)[1]*100,2),'%') print(' ')
gender = Male 的违约概率: 26.16 %
gender = Female 的违约概率: 26.92 %
Partner = No 的违约概率: 32.96 %
Partner = Yes 的违约概率: 19.66 %
Dependents = No 的违约概率: 31.28 %
Dependents = Yes 的违约概率: 15.45 %
PhoneService = Yes 的违约概率: 26.71 %
PhoneService = No 的违约概率: 24.93 %
MultipleLines = No 的违约概率: 25.04 %
MultipleLines = Yes 的违约概率: 28.61 %
MultipleLines = No phone service 的违约概率: 24.93 %
InternetService = Fiber optic 的违约概率: 41.89 %
InternetService = DSL 的违约概率: 18.96 %
InternetService = No 的违约概率: 7.4 %
OnlineSecurity = No 的违约概率: 41.77 %
OnlineSecurity = Yes 的违约概率: 14.61 %
OnlineSecurity = No internet service 的违约概率: 7.4 %
OnlineBackup = No 的违约概率: 39.93 %
OnlineBackup = Yes 的违约概率: 21.53 %
OnlineBackup = No internet service 的违约概率: 7.4 %
DeviceProtection = No 的违约概率: 39.13 %
DeviceProtection = Yes 的违约概率: 22.5 %
DeviceProtection = No internet service 的违约概率: 7.4 %
TechSupport = No 的违约概率: 41.64 %
TechSupport = Yes 的违约概率: 15.17 %
TechSupport = No internet service 的违约概率: 7.4 %
StreamingTV = No 的违约概率: 33.52 %
StreamingTV = Yes 的违约概率: 30.07 %
StreamingTV = No internet service 的违约概率: 7.4 %
StreamingMovies = No 的违约概率: 33.68 %
StreamingMovies = Yes 的违约概率: 29.94 %
StreamingMovies = No internet service 的违约概率: 7.4 %
Contract = Month-to-month 的违约概率: 42.71 %
Contract = Two year 的违约概率: 2.83 %
Contract = One year 的违约概率: 11.27 %
PaperlessBilling = Yes 的违约概率: 33.57 %
PaperlessBilling = No 的违约概率: 16.33 %
PaymentMethod = Electronic check 的违约概率: 45.29 %
PaymentMethod = Mailed check 的违约概率: 19.11 %
PaymentMethod = Bank transfer (automatic) 的违约概率: 16.71 %
PaymentMethod = Credit card (automatic) 的违约概率: 15.24 %
1.2.3单变量可视化（数值型特征）
连续型变量有['SeniorCitizen', 'tenure', 'MonthlyCharges']，再加上上面的 TotalCharges，仔细一卡，SeniorCitizen是否是老年人，这个字段也是离散型的，我们应该将他放到离散型里面，职业有74个值，而且还是加密的，我想把这个变量删除，如果是中文还好处理一点，但是是数值型就有点难处理了
先看下是否老年人这个字段
#先看老年人这个字段 fig, axes = plt.subplots(1,2,figsize=(12,6)) train_data['SeniorCitizen'].value_counts().plot.bar(ax=axes[0]) sns.barplot(train_data['SeniorCitizen'],train_data['Churn'],ax=axes[1]) print(train_data['SeniorCitizen'].value_counts()/len(train_data)) ''' 0 0.837853 1 0.162147 '''
看一下职业的,太离散了，删除吧
再看看两个费用，可知这两个费用相关性还是很高的，有65%多，而且总费用有缺失值'-',
#然后就是 #MonthlyCharges：月费用 #TotalCharges：总费用 #先把总费用转为数值型 train_data.TotalCharges.value_counts() #发现有11个' '值 #反正只有11个值，删除了没有影响 train_data['TotalCharges'].replace(' ',np.nan,inplace=True) train_data=train_data.dropna() train_data.TotalCharges=train_data.TotalCharges.astype(np.float64) train_data.TotalCharges.dtype #dtype('float64') #看一下两个费用相关性如何 train_data[['MonthlyCharges','TotalCharges']].corr() ''' MonthlyCharges TotalCharges MonthlyCharges 1.000000 0.651065 TotalCharges 0.651065 1.000000 '''
看看违约用户的费用的箱型图有啥区别
#看看违约用户的费用的箱型图有啥区别 plt.boxplot((train_data.MonthlyCharges[train_data.Churn==0],train_data.MonthlyCharges[train_data.Churn==1]),labels=('Churn==0','Churn==1')) plt.title('月费用') plt.show() plt.boxplot((train_data.TotalCharges[train_data.Churn==0],train_data.TotalCharges[train_data.Churn==1]),labels=('Churn==0','Churn==1')) plt.title('总费用') plt.show()
可以看得出流水用户的月费用会高一点，但是总费用却低一点，很有可能是月租太贵了，组那么几个月就不用的，我们还可以构建一个使用月长，用总费用除以月费用即是
#费用的分布 fig, axes = plt.subplots(1,2,figsize=(12,6)) sns.distplot( train_data.MonthlyCharges, ax=axes[0]) sns.distplot( train_data.TotalCharges, ax=axes[1])
2. 特征工程
特征工程用的是原数据
train=pd.read_csv('F:\\python\\电信用户数据\\WA_Fn-UseC_-Telco-Customer-Churn.csv')
2.1 缺失值处理
从上面的数据可视化，我们知道总费用有11个‘-’类型的缺失值，我们先使用空值填充，然后再删除
train['TotalCharges'].replace(' ',np.nan,inplace=True) train=train.dropna()
2.2 处理No internet service和No，统一改为NO
处理同一个意思，却有两种表达方式的特征
#处理同一个意思，却有两种表达方式的特征 two_no_col=['MultipleLines' , 'OnlineSecurity', 'OnlineBackup', 'DeviceProtection', 'TechSupport', 'StreamingTV', 'StreamingMovies'] dict1={'No phone service':'No', 'No':'No', 'Yes':'Yes'} train['MultipleLines']=train['MultipleLines'].map(dict1) dict2={'No internet service':'No', 'No':'No', 'Yes':'Yes'} for i in ['OnlineSecurity', 'OnlineBackup', 'DeviceProtection', 'TechSupport', 'StreamingTV', 'StreamingMovies']: train[i]=train[i].map(dict2)
2.3 处理费用
处理费用，首先我们先构造一个使用月长的字段，后面再做两个版本，一个是数值型特征有进行数据分箱，一个是没有分箱
#处理费用，首先我们先构造一个使用月长的字段，后面再做两个版本，一个是数值型特征有进行数据分箱，一个是没有分箱 #先将总费用的类型改为数值型 train.TotalCharges=train.TotalCharges.astype(np.float64) train.TotalCharges.dtype #dtype('float64') #构造使用月长 train['use_month']=round(train.TotalCharges/train_data.MonthlyCharges,0) #特征分箱，这次的版本先不用分箱
2.4 类别特征做One-hot
首先要删除customerID和tenure ：客户的职位，再将需要转化的特征改为str类型，然后再对独热编码，最后将原来的特征删除
#类别特征做One-hot #首先要删除customerID customerID=train.pop('customerID')
train.pop('tenure') train.columns ''' Index(['gender', 'SeniorCitizen', 'Partner', 'Dependents', 'tenure', 'PhoneService', 'MultipleLines', 'InternetService', 'OnlineSecurity', 'OnlineBackup', 'DeviceProtection', 'TechSupport', 'StreamingTV', 'StreamingMovies', 'Contract', 'PaperlessBilling', 'PaymentMethod', 'MonthlyCharges', 'TotalCharges', 'Churn', 'use_month'], dtype='object') ''' onehot_col=['gender', 'SeniorCitizen', 'Partner', 'Dependents', 'tenure', 'PhoneService', 'MultipleLines', 'InternetService', 'OnlineSecurity', 'OnlineBackup', 'DeviceProtection', 'TechSupport', 'StreamingTV', 'StreamingMovies', 'Contract', 'PaperlessBilling', 'PaymentMethod'] train[onehot_col] = train[onehot_col].astype(str) feature_dummies = pd.get_dummies(train[onehot_col]) # feature_dummies.head() train.drop(onehot_col, inplace=True, axis=1) train = pd.concat((train, feature_dummies), axis=1) train.head()
还需要将特征y变成0，1
#首先要将特征值换成0,1 train.Churn[train['Churn']=='No']=0 train.Churn[train['Churn']=='Yes']=1 train.Churn=train.Churn.astype(np.int64)
2.5 画特征相关性的地热图
plt.figure(figsize=(20,16)) sns.heatmap(train.corr(),xticklabels=train.corr().columns, yticklabels=train.corr().columns, linewidths=0.2, cmap="YlGnBu",annot=True)
3. 模型训练
3.1 逻辑回归
首先导入模快和分好x和y
#模型训练 #首先导入模快和分好x和y from sklearn.linear_model import LogisticRegression from sklearn.metrics import confusion_matrix, roc_curve,roc_auc_score,classification_report ,auc from sklearn.model_selection import train_test_split train_y=train['Churn'].values train_x=train x_train,x_test,y_train,y_test=train_test_split(train_x,train_y,test_size=0.3,random_state=0) #逻辑回归 clf = LogisticRegression() clf.fit(x_train.values,y_train) #训练集测试auc p=clf.predict(x_train) fpr, tpr, p_threshold = metrics.roc_curve(y_train, p, drop_intermediate=False, pos_label=1) auc(fpr, tpr) #1 #用测试集进行检验 p=clf.predict(x_test) fpr, tpr, p_threshold = metrics.roc_curve(y_test, p, drop_intermediate=False, pos_label=1) auc(fpr, tpr) #1
是特征处理的太好了吗，为什么结果会是1 ，太奇怪了
3.2 Random Forest
from sklearn.ensemble import RandomForestClassifier from sklearn.model_selection import cross_val_score import sklearn %matplotlib inline depth_ = [1, 2, 3, 4, 5, 6, 7, 8] scores = [] for depth in depth_: clf1 = RandomForestClassifier(n_estimators=100, max_depth=depth, random_state=0) test_score = cross_val_score(clf1, x_train.values,y_train, cv=10, scoring="precision") scores.append(np.mean(test_score)) plt.plot(depth_, scores)
结果也是很不错，看来是特征工程太完美了
3.3 XGBClassifier
from xgboost import XGBClassifier params = [1, 2, 3, 4, 5, 6] test_scores = [] for param in params: clf = XGBClassifier(max_depth=param) test_score = cross_val_score(clf, x_train.values,y_train, cv=10, scoring="precision") test_scores.append(np.mean(test_score)) plt.plot(params, test_scores)
天哪，我自闭了，还是1。
附上全部代码
# 模块 import pandas as pd import numpy as np import matplotlib.pyplot as plt import seaborn as sns import scipy.stats as st plt.rc("font",family="SimHei",size="12") #解决中文无法显示的问题 # 导入数据 dianxin=pd.read_csv('F:\\python\\电信用户数据\\WA_Fn-UseC_-Telco-Customer-Churn.csv') train_data.head() train_data.describe() train_data.info() train_data.shape #(7043, 21) #类别特征 object_col=dianxin.select_dtypes(include=[np.object]).columns ''' Index(['customerID', 'gender', 'Partner', 'Dependents', 'PhoneService', 'MultipleLines', 'InternetService', 'OnlineSecurity', 'OnlineBackup', 'DeviceProtection', 'TechSupport', 'StreamingTV', 'StreamingMovies', 'Contract', 'PaperlessBilling', 'PaymentMethod', 'TotalCharges', 'Churn'], dtype='object') ''' #数值特征 num_col=dianxin.select_dtypes(include=[np.number]).columns ''' Index(['SeniorCitizen', 'tenure', 'MonthlyCharges'], dtype='object') ''' #单变量可视化 #首先看一些大体的分布如何，后面才好画图 for i in object_col: print(i,train_data[i].nunique()) #customerID，TotalCharges,可以不用画图 object_col=['gender', 'Partner', 'Dependents', 'PhoneService', 'MultipleLines', 'InternetService', 'OnlineSecurity', 'OnlineBackup', 'DeviceProtection', 'TechSupport', 'StreamingTV', 'StreamingMovies', 'Contract', 'PaperlessBilling', 'PaymentMethod', 'Churn'] fig = plt.figure(figsize=[16,40]) for col,i in zip(object_col,range(1,17)): axes = fig.add_subplot(9,2,i) sns.countplot(train_data[col],ax=axes) plt.show() for i in object_col: for j in train_data[i].value_counts(normalize = True).index: print(i,'=',j,":",round(train_data[i].value_counts(normalize = True)[j]*100,2),'%') print(' ') ''' 1.样本是否均匀 gender：性别。（Female & Male） 比较平均 Partner ：是否有配偶 （Yes or No）比较平均 Dependents ：是否经济独立 （Yes or No）不独立的占比约60% PhoneService ：是否开通电话服务业务 （Yes or No） 很不均衡，开通的约占80%多 MultipleLines：是否开通了多线业务（Yes 、No or No phoneservice 三种） InternetService：是否开通互联网服务 （No, DSL数字网络，fiber optic光纤网络 三种） OnlineSecurity：是否开通网络安全服务（Yes，No，No internetserive 三种） OnlineBackup：是否开通在线备份业务（Yes，No，No internetserive 三种） DeviceProtection：是否开通了设备保护业务（Yes，No，No internetserive 三种） TechSupport：是否开通了技术支持服务（Yes，No，No internetserive 三种） StreamingTV：是否开通网络电视（Yes，No，No internetserive 三种） StreamingMovies：是否开通网络电影（Yes，No，No internetserive 三种） Contract：签订合同方式 （按月，一年，两年） PaperlessBilling：是否开通电子账单（Yes or No） PaymentMethod：付款方式（bank transfer，credit card，electronic check，mailed check） Churn：该用户是否流失（Yes or No） ''' #但是我们注意到下面，有三个值（类似于No phone service和No）其实是二个值 two_no_col=['MultipleLines' , 'OnlineSecurity', 'OnlineBackup', 'DeviceProtection', 'TechSupport', 'StreamingTV', 'StreamingMovies'] dict1={'No phone service':'No', 'No':'No', 'Yes':'Yes'} train_data['MultipleLines']=train_data['MultipleLines'].map(dict1) dict2={'No internet service':'No', 'No':'No', 'Yes':'Yes'} for i in ['OnlineSecurity', 'OnlineBackup', 'DeviceProtection', 'TechSupport', 'StreamingTV', 'StreamingMovies']: train_data[i]=train_data[i].map(dict2) fig = plt.figure(figsize=[16,20]) for col,i in zip(two_no_col,range(1,8)): axes = fig.add_subplot(4,2,i) sns.countplot(train_data[col],ax=axes) plt.show() for i in two_no_col: for j in train_data[i].value_counts(normalize = True).index: print(i,'=',j,":",round(train_data[i].value_counts(normalize = True)[j]*100,2),'%') print(' ') #类别变量和y值的可视化 #但是要注意，barplot需要y值是数值型的，现在我们先把y值转化成数值型的，Churn=1即是逾期 train_data.Churn[train_data['Churn']=='No']=0 train_data.Churn[train_data['Churn']=='Yes']=1 fig = plt.figure(figsize=[16,40]) for col,i in zip(object_col,range(1,17)): axes = fig.add_subplot(8,2,i) sns.barplot(train_data[col],train_data['Churn'],ax=axes) plt.show() #不要y值 col=['gender', 'Partner', 'Dependents', 'PhoneService', 'MultipleLines', 'InternetService', 'OnlineSecurity', 'OnlineBackup', 'DeviceProtection', 'TechSupport', 'StreamingTV', 'StreamingMovies', 'Contract', 'PaperlessBilling', 'PaymentMethod',] for i in col: for j in train_data[i].value_counts(normalize = True).index: print(i,'=',j,"的违约概率:",round(train_data.Churn[train_data[i]==j].value_counts(normalize = True)[1]*100,2),'%') print(' ') #单变量可视化（连续型） #['SeniorCitizen', 'tenure', 'MonthlyCharges'] #再加上上面的 TotalCharges #仔细一卡，SeniorCitizen是否是老年人，这个字段也是离散型的，我们应该将他放到离散型里面， #职业有74个值，而且还是加密的，我想把这个变量删除，如果是中文还好处理一点，但是是数值型就有点难处理了 num_col=['SeniorCitizen', 'tenure', 'MonthlyCharges','TotalCharges'] train_data[num_col] #先看老年人这个字段 fig, axes = plt.subplots(1,2,figsize=(12,6)) train_data['SeniorCitizen'].value_counts().plot.bar(ax=axes[0]) sns.barplot(train_data['SeniorCitizen'],train_data['Churn'],ax=axes[1]) print(train_data['SeniorCitizen'].value_counts()/len(train_data)) #看一下职业的,太离散了，删除吧 #然后就是 #MonthlyCharges：月费用 #TotalCharges：总费用 #先把总费用转为数值型 train_data.TotalCharges.value_counts() #发现有11个' '值 #反正只有11个值，删除了没有影响 train_data['TotalCharges'].replace(' ',np.nan,inplace=True) train_data=train_data.dropna() train_data.TotalCharges=train_data.TotalCharges.astype(np.float64) train_data.TotalCharges.dtype #dtype('float64') #看一下两个费用相关性如何 train_data[['MonthlyCharges','TotalCharges']].corr() ''' MonthlyCharges TotalCharges MonthlyCharges 1.000000 0.651065 TotalCharges 0.651065 1.000000 ''' #看看违约用户的费用的箱型图有啥区别 plt.boxplot((train_data.MonthlyCharges[train_data.Churn==0],train_data.MonthlyCharges[train_data.Churn==1]),labels=('Churn==0','Churn==1')) plt.title('月费用') plt.show() plt.boxplot((train_data.TotalCharges[train_data.Churn==0],train_data.TotalCharges[train_data.Churn==1]),labels=('Churn==0','Churn==1')) plt.title('总费用') plt.show() #费用的分布 fig, axes = plt.subplots(1,2,figsize=(12,6)) sns.distplot( train_data.MonthlyCharges, ax=axes[0]) sns.distplot( train_data.TotalCharges, ax=axes[1]) #特征工程 train=pd.read_csv('F:\\python\\电信用户数据\\WA_Fn-UseC_-Telco-Customer-Churn.csv') #1.缺失值处理,我们知道总费用有11个‘-’类型的缺失值，我们先使用空值填充，然后再删除 train['TotalCharges'].replace(' ',np.nan,inplace=True) train=train.dropna() #处理同一个意思，却有两种表达方式的特征 two_no_col=['MultipleLines' , 'OnlineSecurity', 'OnlineBackup', 'DeviceProtection', 'TechSupport', 'StreamingTV', 'StreamingMovies'] dict1={'No phone service':'No', 'No':'No', 'Yes':'Yes'} train['MultipleLines']=train['MultipleLines'].map(dict1) dict2={'No internet service':'No', 'No':'No', 'Yes':'Yes'} for i in ['OnlineSecurity', 'OnlineBackup', 'DeviceProtection', 'TechSupport', 'StreamingTV', 'StreamingMovies']: train[i]=train[i].map(dict2) #处理费用，首先我们先构造一个使用月长的字段，后面再做两个版本，一个是数值型特征有进行数据分箱，一个是没有分箱 #先将总费用的类型改为数值型 train.TotalCharges=train.TotalCharges.astype(np.float64) train.TotalCharges.dtype #dtype('float64') #构造使用月长 train['use_month']=round(train.TotalCharges/train_data.MonthlyCharges,0) #特征分箱，这次的版本先不用分箱 #类别特征做One-hot #首先要删除customerID customerID=train.pop('customerID') train.pop('tenure') train.columns ''' Index(['gender', 'SeniorCitizen', 'Partner', 'Dependents', 'tenure', 'PhoneService', 'MultipleLines', 'InternetService', 'OnlineSecurity', 'OnlineBackup', 'DeviceProtection', 'TechSupport', 'StreamingTV', 'StreamingMovies', 'Contract', 'PaperlessBilling', 'PaymentMethod', 'MonthlyCharges', 'TotalCharges', 'Churn', 'use_month'], dtype='object') ''' onehot_col=['gender', 'SeniorCitizen', 'Partner', 'Dependents', 'PhoneService', 'MultipleLines', 'InternetService', 'OnlineSecurity', 'OnlineBackup', 'DeviceProtection', 'TechSupport', 'StreamingTV', 'StreamingMovies', 'Contract', 'PaperlessBilling', 'PaymentMethod'] train[onehot_col] = train[onehot_col].astype(str) feature_dummies = pd.get_dummies(train[onehot_col]) # feature_dummies.head() train.drop(onehot_col, inplace=True, axis=1) train = pd.concat((train, feature_dummies), axis=1) train.head() #首先要将特征值换成0,1 train.Churn[train['Churn']=='No']=0 train.Churn[train['Churn']=='Yes']=1 train.Churn=train.Churn.astype(np.int64) plt.figure(figsize=(20,16)) sns.heatmap(train.corr(),xticklabels=train.corr().columns, yticklabels=train.corr().columns, linewidths=0.2, cmap="YlGnBu",annot=True) #模型训练 #首先导入模快和分好x和y from sklearn.linear_model import LogisticRegression from sklearn.metrics import confusion_matrix, roc_curve,roc_auc_score,classification_report ,auc from sklearn.model_selection import train_test_split train_y=train['Churn'].values train_x=train x_train,x_test,y_train,y_test=train_test_split(train_x,train_y,test_size=0.3,random_state=0) #逻辑回归 clf = LogisticRegression() clf.fit(x_train.values,y_train) #训练集测试auc p=clf.predict(x_train) fpr, tpr, p_threshold = metrics.roc_curve(y_train, p, drop_intermediate=False, pos_label=1) auc(fpr, tpr) #1 #用测试集进行检验 p=clf.predict(x_test) fpr, tpr, p_threshold = metrics.roc_curve(y_test, p, drop_intermediate=False, pos_label=1) auc(fpr, tpr) #1 ######################################################################### D
【推荐】国内首个AI IDE，深度理解中文开发场景，立即下载体验Trae
【推荐】编程新体验，更懂你的AI，立即体验豆包MarsCode编程助手
【推荐】抖音旗下AI助手豆包，你的智能百科全书，全免费不限次数
【推荐】轻量又高性能的 SSH 工具 IShell：AI 加持，快人一步
· AI与.NET技术实操系列：向量存储与相似性搜索在 .NET 中的实现
· 基于Microsoft.Extensions.AI核心库实现RAG应用
· Linux系列：如何用heaptrack跟踪.NET程序的非托管内存泄露
· 开发者必知的日志记录最佳实践
· SQL Server 2025 AI相关能力初探
· winform 绘制太阳，地球，月球 运作规律
· 震惊！C++程序真的从main开始吗？99%的程序员都答错了
· AI与.NET技术实操系列（五）：向量存储与相似性搜索在 .NET 中的实现
· 【硬核科普】Trae如何「偷看」你的代码？零基础破解AI编程运行原理
· 超详细：普通电脑也行Windows部署deepseek R1训练数据并当服务器共享给他人