本文基于LightGBM实现银行客户信用违约预测。先读取训练集和测试集数据，处理Credit_Product列缺失值为“ No ”。经数据EDA后，用LabelEncoder对离散变量编码。构建LGBM模型，通过5折交叉验证训练，验证集平均AUC为0.7889，最后输出预测结果为指定格式的CSV文件。
☞☞☞AI 智能聊天, 问答助手, AI 智能搜索, 免费无限量使用 DeepSeek R1 模型☜☜☜
题目地址：Coggle竞赛
信用评分卡（金融风控）是金融行业和通讯行业常见的风控手段，通过对客户提交的个人信息和数据来预测未来违约的可能性。对客户进行信用评分是一个常见的分类问题。
在本次赛题中需要参赛选手建立机器学习模型来预测申请人是“好”还是“坏”客户，与其他任务不同，没有给出“好”或“坏”的定义。 您应该使用一些技术，例如年份分析来构建您的标签。
赛题包含两部分训练集和测试集，选手需要在训练集上进行搭建模型，然后在测试集进行预测。
数据字段介绍如下：
评分使用准确率进行评分，准确率值越大越好。
提交格式样例：
ID,TargetAXM2EH3R,18ETNJAUW,1VCSJTEPW,09EOYOOHV,0
学习自：
通过pandas读取数据
import pandas as pdimport numpy as np
df=pd.read_csv("data/data207852/train.csv")
test=pd.read_csv("data/data207852/test.csv")
test.head(10)ID Gender Age Region_Code Occupation Channel_Code Vintage \ 0 AXM2EH3R Female 43 RG284 Self_Employed X3 26 1 8ETNJAUW Female 46 RG282 Self_Employed X2 14 2 VCSJTEPW Female 28 RG254 Self_Employed X1 15 3 9EOYOOHV Male 58 RG265 Other X3 15 4 S4B53OKJ Male 75 RG260 Other X3 111 5 3DTSVD9Y Female 51 RG268 Self_Employed X1 57 6 8WYWQUUX Male 32 RG279 Salaried X1 33 7 FPQTNHGY Female 38 RG270 Salaried X1 33 8 UXCKDQ34 Male 56 RG254 Self_Employed X2 62 9 CFTGOZHH Female 29 RG283 Salaried X1 20 Credit_Product Avg_Account_Balance Is_Active 0 Yes 1325325 Yes 1 No 634489 No 2 No 2215655 No 3 Yes 925929 Yes 4 No 721825 Yes 5 No 490345 No 6 No 650483 No 7 NaN 369777 No 8 Yes 2406880 Yes 9 No 659053 No
df.head(10)
ID Gender Age Region_Code Occupation Channel_Code Vintage \ 0 ZYFGCP3R Male 58 RG264 Self_Employed X2 19 1 MQJBCRCF Female 45 RG271 Self_Employed X3 104 2 UZOQRG46 Female 30 RG278 Other X1 25 3 GCX6RVZS Female 52 RG283 Self_Employed X1 43 4 9V6BRARI Female 76 RG254 Other X1 57 5 WUGN99OM Male 28 RG275 Salaried X1 33 6 EQ4CBNED Male 31 RG268 Salaried X1 33 7 JZZ7MPIR Male 48 RG259 Entrepreneur X2 67 8 KVHMRSES Female 31 RG254 Salaried X1 33 9 KS45GJCT Female 48 RG273 Other X3 105 Credit_Product Avg_Account_Balance Is_Active Target 0 No 552449 Yes 0 1 Yes 525206 No 1 2 No 724718 No 0 3 Yes 1452453 No 0 4 No 1895762 No 0 5 No 885576 No 0 6 No 653135 Yes 0 7 Yes 389553 Yes 1 8 No 1543001 No 0 9 NaN 360005 Yes 1
df.info()
<class 'pandas.core.frame.DataFrame'> RangeIndex: 195725 entries, 0 to 195724 Data columns (total 11 columns): # Column Non-Null Count Dtype --- ------ -------------- ----- 0 ID 195725 non-null object 1 Gender 195725 non-null object 2 Age 195725 non-null int64 3 Region_Code 195725 non-null object 4 Occupation 195725 non-null object 5 Channel_Code 195725 non-null object 6 Vintage 195725 non-null int64 7 Credit_Product 172279 non-null object 8 Avg_Account_Balance 195725 non-null int64 9 Is_Active 195725 non-null object 10 Target 195725 non-null int64 dtypes: int64(4), object(7) memory usage: 16.4+ MB
发现Credit_Product列有空值，怎么办？仔细一看test也有空值，不能简单的删除了，那就看这个值哪个多就填写哪个了。
# 统计某列值 df['Credit_Product'].unique()
array(['No', 'Yes', nan], dtype=object)
# 统计某列出现某值的次数 df['Credit_Product'].value_counts()
No 114910 Yes 57369 Name: Credit_Product, dtype: int64
可以看出，该列值主要为No，因此缺失值nan设置为No。
test.info()
<class 'pandas.core.frame.DataFrame'> RangeIndex: 50000 entries, 0 to 49999 Data columns (total 10 columns): # Column Non-Null Count Dtype --- ------ -------------- ----- 0 ID 50000 non-null object 1 Gender 50000 non-null object 2 Age 50000 non-null int64 3 Region_Code 50000 non-null object 4 Occupation 50000 non-null object 5 Channel_Code 50000 non-null object 6 Vintage 50000 non-null int64 7 Credit_Product 44121 non-null object 8 Avg_Account_Balance 50000 non-null int64 9 Is_Active 50000 non-null object dtypes: int64(3), object(7) memory usage: 3.8+ MB
# 空值填Nodf=df.fillna('No')
test=test.fillna('No')
import matplotlib.pyplot as plt
import seaborn as sns
%matplotlib inline# 按年龄分布查看ages=[22,30,40,50,60,70,80,90]
df1=df[df['Credit_Product']=='Yes']
binning=pd.cut(df1['Age'],ages,right=False)
time=pd.value_counts(binning)# 可视化time=time.sort_index()
fig=plt.figure(figsize=(6,2),dpi=120)
sns.barplot(time.index,time,color='royalblue')
x=np.arange(len(time))
y=time.valuesfor x_loc,jobs in zip(x,y):
plt.text(x_loc, jobs+2, '{:.1f}%'.format(jobs/sum(time)*100), ha='center', va= 'bottom',fontsize=8)
plt.xticks(fontsize=8)
plt.yticks([])
plt.ylabel('')
plt.title('duration_yes',size=8)
sns.despine(left=True)
plt.show()<Figure size 720x240 with 1 Axes>
# 分离数值变量与分类变量Nu_feature = list(df.select_dtypes(exclude=['object']).columns)
Ca_feature = list(df.select_dtypes(include=['object']).columns)#查看训练集与测试集数值变量分布import matplotlib.pyplot as plt
import seaborn as snsimport warnings
warnings.filterwarnings("ignore")
plt.figure(figsize=(15,5))
Nu_feature.remove('Target')# 根据数值型分布查看i=1for col in Nu_feature:
ax=plt.subplot(1,3,i)
ax=sns.kdeplot(df[col],color='red')
ax=sns.kdeplot(test[col],color='cyan')
ax.set_xlabel(col)
ax.set_ylabel('Frequency')
ax=ax.legend(['train','test'])
i+=1plt.show()<Figure size 1500x500 with 3 Axes>
查看离散变量分布
由于时间太久，故不测
col1=Ca_feature
plt.figure(figsize=(20,10))j=1for col in col1: ax=plt.subplot(6,3,j) ax=plt.scatter(x=range(len(df)),y=df[col],color='red')
plt.title(col)
j+=1k=7for col in col1: ax=plt.subplot(6,3,k) ax=plt.scatter(x=range(len(test)),y=test[col],color='cyan')
plt.title(col)
k+=1
plt.subplots_adjust(wspace=0.4,hspace=0.3)
plt.show()# 离散数据Encoderfrom sklearn.preprocessing import LabelEncoder
lb = LabelEncoder()
cols = Ca_featurefor m in cols:
df[m] = lb.fit_transform(df[m])
test[m] = lb.fit_transform(test[m])
correlation_matrix=df.corr()
plt.figure(figsize=(12,10))# 热力图sns.heatmap(correlation_matrix,vmax=0.9,linewidths=0.05,cmap="RdGy")<matplotlib.axes._subplots.AxesSubplot at 0x7fa254ac6150>
<Figure size 1200x1000 with 2 Axes>
这里使用留出法划分数据集，将数据集分为自变量和因变量。
按比例切割训练集和测试集(一般测试集的比例有30%、25%、20%、15%和10%)，使用分层抽样，设置随机种子以便结果能复现
from lightgbm.sklearn import LGBMClassifierfrom sklearn.model_selection import train_test_splitfrom sklearn.model_selection import KFoldfrom sklearn.metrics import accuracy_score, auc, roc_auc_score X=df.drop(columns=['ID','Target']) Y=df['Target'] test=test.drop(columns='ID')# 划分训练及测试集x_train,x_test,y_train,y_test = train_test_split( X, Y,test_size=0.3,random_state=1)
创建基于树的分类模型（lightgbm）
这些模型进行训练，分别的到训练集和测试集的得分
# 建立模型gbm = LGBMClassifier(n_estimators=600,learning_rate=0.01,boosting_type= 'gbdt',
objective = 'binary',
max_depth = -1,
random_state=2022,
metric='auc')交叉验证介绍
# 交叉验证result1 = []
mean_score1 = 0n_folds=5kf = KFold(n_splits=n_folds ,shuffle=True,random_state=2022)for train_index, test_index in kf.split(X):
x_train = X.iloc[train_index]
y_train = Y.iloc[train_index]
x_test = X.iloc[test_index]
y_test = Y.iloc[test_index]
gbm.fit(x_train,y_train)
y_pred1=gbm.predict_proba((x_test),num_iteration=gbm.best_iteration_)[:,1] print('验证集AUC:{}'.format(roc_auc_score(y_test,y_pred1)))
mean_score1 += roc_auc_score(y_test,y_pred1)/ n_folds
y_pred_final1 = gbm.predict_proba((test),num_iteration=gbm.best_iteration_)[:,1]
y_pred_test1=y_pred_final1
result1.append(y_pred_test1)验证集AUC:0.7889931707362382 验证集AUC:0.7894677985120346 验证集AUC:0.7931272562656144 验证集AUC:0.7850546301430752 验证集AUC:0.7876841341097264
# 模型评估print('mean 验证集auc:{}'.format(mean_score1))
cat_pre1=sum(result1)/n_foldsmean 验证集auc:0.7888653979533378
将预测结果按照指定格式输出到result.csv文件中
ret1=pd.DataFrame(cat_pre1,columns=['Target'])
ret1['Target']=np.where(ret1['Target']>0.5,'1','0').astype('str')
result = pd.DataFrame()
test=pd.read_csv("data/data207852/test.csv")
result['ID'] = test['ID']
result['Target'] = ret1['Target']
result.to_csv('result.csv',index=False)print(test.columns)
Index(['ID', 'Gender', 'Age', 'Region_Code', 'Occupation', 'Channel_Code',
'Vintage', 'Credit_Product', 'Avg_Account_Balance', 'Is_Active'],
dtype='object')第一次提交错了，第二次刷新过头了
以上就是基于LightGBM实现银行客户信用违约预测的详细内容，更多请关注php中文网其它相关文章！
Copyright 2014-2025 https://www.php.cn/ All Rights Reserved | php.cn | 湘ICP备2023035733号
999
收藏
