下采样和SMOTE算法（过采样）建模评估信用卡风险比较

1.导入相关库

import pandas as pd
import matplotlib.pyplot as plt
from pylab import mpl
import numpy as np

2.数据预处理

共284807条数据

3.绘制图形，查看正负样本个数

mpl.rcParams['font.sans-serif']=['Microsoft YaHei']
mpl.rcParams['axes.unicode_minus']=False
label_count=pd.value_counts(data['Class'])
plt.title("正负例样本数")
plt.xlabel("类别")
plt.ylabel("频数")
label_count.plot(kind='bar')
plt.show()

                                          

4.采用Z标准化对Amount进行数据标准化

from sklearn.preprocessing import StandardScaler
scaler=StandardScaler()
data['Amount']=scaler.fit_transform(data[['Amount']])
data.head()

                                                                        

5.删除无用例

data=data.drop(['Time'],axis=1)

6.逻辑回归

6.1下采样解决样本不均衡

6.1.1分离对立样本

positive_eg=data[data['Class']==0]
negative_eg=data[data['Class']==1]

 

6.1.2从0样本随机选取492条数据并进行拼接

np.random.seed(seed=2)
positive_eg=positive_eg.sample(len(negative_eg))
data_c=pd.concat([positive_eg,negative_eg])

6.2下采样数据切分

from sklearn.model_selection import train_test_split
X=data_c.drop('Class',axis=1)
y=data_c.Class
x_train,x_test,y_train,y_test=\
   train_test_split(X,y,test_size=0.3,random_state=0)

 6.3对原始数据切分

X_whole=data.drop('Class',axis=1)
y_whole=data.Class
x_train_w,x_test_w,y_train_w,y_test_w=\
   train_test_split(X_whole,y_whole,test_size=0.3,random_state=0)

6.4执行交叉验证选择最优惩罚因子

from sklearn.linear_model import LogisticRegression
from sklearn.model_selection import cross_val_score
scores = []
c_param_range = [0.01,0.1,1,10,100]
z = 1
for i in c_param_range:
    start_time = time.time()
    lr = LogisticRegression(C = i, penalty = 'l2', solver='lbfgs')
    score = cross_val_score(lr, x_train, y_train, cv=10, scoring='recall')
    score_mean = sum(score)/len(score)
    scores.append(score_mean)
    end_time = time.time()
    print("第{}次...".format(z))
    print("time spend:{:.2f}".format(end_time - start_time))
    print("recall值为:{}".format(score_mean))
    z +=1

best_c = c_param_range[np.argmax(scores)]
print()
print("最优惩罚因子为: {}".format(best_c))

 

6.5采用最优惩罚因子建立最优模型进行训练

lr=LogisticRegression(C=best_c,penalty='l2')
lr.fit(x_train,y_train)

6.6预测结果

train_predicted=lr.predict(x_train)

6.7绘制混淆矩阵

from sklearn import metrics
print(metrics.classification_report(y_train,train_predicted))

相当不错的召回率

6.7.1可视化混淆矩阵

def cm_plot(y, yp):
    from sklearn.metrics import confusion_matrix
    import matplotlib.pyplot as plt

    cm = confusion_matrix(y, yp)
    plt.matshow(cm, cmap=plt.cm.Blues)
    plt.colorbar()
    for x in range(len(cm)):
        for y in range(len(cm)):
            plt.annotate(cm[x, y], xy=(y, x), horizontalalignment='center',
                         verticalalignment='center')
            plt.ylabel('True label')
            plt.xlabel('Predicted label')
    return plt

cm_plot(y_train,train_predicted).show()

 

6.8使用大数据测试集进行测试

test_predicted=lr.predict(x_test_w)
print(metrics.classification_report(y_test_w,test_predicted))
cm_plot(y_test_w,test_predicted).show()

测试效果较好

SMOTE算法

import pandas as pd
import matplotlib.pyplot as plt
from pylab import mpl
import numpy as np
import time

data = pd.read_csv(r".creditcard.csv",\
                   encoding='utf8',engine = 'python')
data.head()

from sklearn.preprocessing import StandardScaler
scaler=StandardScaler()
data['Amount']=scaler.fit_transform(data[['Amount']])
data.head()

data=data.drop(['Time'],axis=1)


from sklearn.model_selection import train_test_split

X_whole=data.drop('Class',axis=1)
y_whole=data.Class
x_train_w,x_test_w,y_train_w,y_test_w=\
   train_test_split(X_whole,y_whole,test_size=0.3,random_state=0)

from imblearn.over_sampling import SMOTE
oversampler=SMOTE(random_state=0)
os_x_train,os_y_train=oversampler.fit_sample(x_train_w,y_train_w)
os_x_train=pd.DataFrame(os_x_train)
os_y_train=pd.Series(os_y_train)

from sklearn.linear_model import LogisticRegression
from sklearn.model_selection import cross_val_score
scores = []
c_param_range = [0.01,0.1,1,10,100]
z = 1
for i in c_param_range:
    start_time = time.time()
    lr = LogisticRegression(C = i, penalty = 'l2', solver='lbfgs')
    score = cross_val_score(lr, os_x_train, os_y_train, cv=10, scoring='roc_auc')
    score_mean = sum(score)/len(score)
    scores.append(score_mean)
    end_time = time.time()
    print("第{}次...".format(z))
    print("time spend:{:.2f}".format(end_time - start_time))
    print("recall值为:{}".format(score_mean))
    z +=1

best_c = c_param_range[np.argmax(scores)]

lr=LogisticRegression(C=best_c,penalty='l2')
lr.fit(os_x_train,os_y_train)

train_predicted=lr.predict(os_x_train)

from sklearn import metrics
print(metrics.classification_report(os_y_train,train_predicted))


def cm_plot(y, yp):
    from sklearn.metrics import confusion_matrix
    import matplotlib.pyplot as plt

    cm = confusion_matrix(y, yp)
    plt.matshow(cm, cmap=plt.cm.Blues)
    plt.colorbar()
    for x in range(len(cm)):
        for y in range(len(cm)):
            plt.annotate(cm[x, y], xy=(y, x), horizontalalignment='center',
                         verticalalignment='center')
            plt.ylabel('True label')
            plt.xlabel('Predicted label')
    return plt

cm_plot(os_y_train,train_predicted).show()

test_predicted=lr.predict(x_test_w)
print(metrics.classification_report(y_test_w,test_predicted))
cm_plot(y_test_w,test_predicted).show()

训练集和大数据的测试集预测效果均比较好

 过采样方法的数据训练更多，相比较来说更好一些。