前言:一个人的信用评级一般用人物画像来评判,如何从很多的人物特征中提取有用的特征呢?
下面以一个金融反欺诈模型为例子来对特征提取有一个简单的理解。
数据下载地址:Notes offered by Prospectus (https://www.lendingclub.com/info/prospectus.action)
一共有145行特征,
1删除了肉眼看的见的空值列
import pandas as pd
import numpy as np
import sys
df = pd.read_csv('./data/LoanStats3a.csv', skiprows = 1, low_memory = True)#skiprows跳过第一行,low_memory低内存加载,报错就该成False
'''读入接待信息'''
# print(df.head(10))
# print(df.info())
'''查看数据特征表格信息'''
df.drop('id', axis = 1, inplace = True)
df.drop('member_id', axis = 1, inplace = True)
2清洗数据,去除特征中的特殊字符
df.term.replace(to_replace = '[^0-9]+', value = '', inplace = True, regex = True)#regex正则打开
df.int_rate.replace('%', value = '', inplace = True)#去不掉说明就是浮点型
df.drop('sub_grade', axis = 1, inplace = True)
df.drop('emp_title', axis = 1, inplace = True)
df.emp_length.replace('n/a', np.nan, inplace = True)
df.emp_length.replace(to_replace = '[^0-9]+', value = '', inplace = True, regex = True)
#这一步是必须做的,这样做以后才能,用info查看
df.dropna(axis = 1, how = 'all', inplace = True)
df.dropna(axis = 0, how = 'all', inplace = True)
3删除空值较多的列
'''debt_settlement_flag_date 98 non-null object
settlement_status 155 non-null object
settlement_date 155 non-null object
settlement_amount 155 non-null float64
settlement_percentage 155 non-null float64
settlement_term 155 non-null float64'''
df.drop(['debt_settlement_flag_date','settlement_status','settlement_date',\
'settlement_amount','settlement_percentage',\
'settlement_term'], axis = 1, inplace = True)
4删除不为空,但是重复较多的列;先删float,再删object
# for col in df.select_dtypes(include = ['float']).columns:
# print('col {} has {}'.format(col, len(df[col].unique())))
'''
col delinq_2yrs has 13
col inq_last_6mths has 29
col mths_since_last_delinq has 96
col mths_since_last_record has 114
col open_acc has 45
col pub_rec has 7
col total_acc has 84
col out_prncp has 1
col out_prncp_inv has 1
col collections_12_mths_ex_med has 2
col policy_code has 1
col acc_now_delinq has 3
col chargeoff_within_12_mths has 2
col delinq_amnt has 4
col pub_rec_bankruptcies has 4
col tax_liens has 3
'''
df.drop(['delinq_2yrs','inq_last_6mths','mths_since_last_delinq',\
'mths_since_last_record','open_acc','pub_rec','total_acc',\
'out_prncp','out_prncp_inv','collections_12_mths_ex_med',\
'policy_code','acc_now_delinq','chargeoff_within_12_mths',\
'delinq_amnt','pub_rec_bankruptcies',\
'tax_liens'], axis = 1, inplace = True)
'''删除objetct类型中数据重复较多的值'''
# for col in df.select_dtypes(include = ['object']).columns:
# print('col {} has {}'.format(col, len(df[col].unique())))
'''
col term has 2
col grade has 7
col emp_length has 11
col home_ownership has 5
col verification_status has 3
col issue_d has 55
col pymnt_plan has 1
col purpose has 1
col zip_code has 837
col addr_state has 50
col earliest_cr_line has 531
col initial_list_status has 1
col last_pymnt_d has 113
col next_pymnt_d has 99
col last_credit_pull_d has 125
col application_type has 1
col hardship_flag has 1
col disbursement_method has 1
col debt_settlement_flag has 2
'''
df.drop(['term','grade','emp_length','home_ownership','verification_status'\
,'issue_d','pymnt_plan','purpose','zip_code','addr_state',\
'earliest_cr_line','initial_list_status','last_pymnt_d',\
'next_pymnt_d','last_credit_pull_d','application_type','hardship_flag',
'disbursement_method','debt_settlement_flag'], axis = 1, inplace = True)
df.drop(['desc','title'], axis = 1, inplace = True)
5标签二值化
df.loan_status.replace('Fully Paid', value = int(1), inplace = True)
df.loan_status.replace('Charged Off', value = int(0), inplace = True)
df.loan_status.replace('Does not meet the credit policy. Status:Fully Paid', \
np.nan, inplace = True)
df.loan_status.replace('Does not meet the credit policy. Status:Charged Off', \
np.nan, inplace = True)
'''删除标签为空的实力,大概删除了3000个不到的实力'''
df.dropna(subset = ['loan_status'], how = 'any', inplace = True)
6把样本中的空值用0.0去填充
df.fillna(0.0, inplace = True)
7计算清洁后样本数据的相关性,删除相关系数大于0.95的列
cor = df.corr()#协方差矩阵
# cor.iloc[:, :] = np.tril(cor, k= -1)
# cor = cor.stack()
# print(cor[(cor>0.55)|(cor<-0.55)])
# sys.exit(0)
'''loan_amnt
funded_amnt
total_pymnt'''
'''删除相关系数大于0.95的列'''
df.drop(['loan_amnt','funded_amnt','total_pymnt'], axis = 1, inplace = True)
print(df.info())#revol_util 39786 non-null object"%"会默认为object,其实他是数值
# sys.exit(0)
df = pd.get_dummies(df)#哑变量
df.to_csv('./data/feature03.csv')