[人工智能]贷款申请最大化利润-机器学习项目实战

一、项目介绍

本项目以某互联网贷款网站提供的贷款人的个人信息为背景，根据历史数据，建立模型，预测新的一个人来了，是否给他贷款，以实现利润最大化。
网站地址：http://lendingclub.com
数据：链接：https://pan.baidu.com/s/1NoU9oGnRS70d8663SQPKVQ
提取码：oldg

二、数据清洗过滤无用特征

#去掉空值太多的行,dropna(thresh=n)非nan最少n个才能保留
#去掉“描述”“链接”等没用的行
#保存为新文件
import pandas as pd
loans2007=pd.read_csv(r"E:\BaiduNetdiskDownload\唐宇迪机器学习\贷款利润最大化\LoanStats3a.csv",engine="python",skiprows=1)#因文件中含有中文，所以应该指定engine为python，否则会报错 OSError: Initializing from file failed;并且药剂得添加参数skiprows=1,否则会报错 KeyError: "labels ['desc' 'url'] not contained in axis"
half=len(loans2007)/2
loans2007=loans2007.dropna(thresh=half,axis=1)
#loans2007=loans2007.drop(["desc","url"],axis=1)
loans2007= loans2007.drop(['desc', 'url'],axis=1)
loans2007.to_csv("E:\BaiduNetdiskDownload\唐宇迪机器学习\贷款利润最大化\loans2007.csv",index=False)

#打印第一列数据
#打印列数
loans2007=pd.read_csv("E:\BaiduNetdiskDownload\唐宇迪机器学习\贷款利润最大化\loans2007.csv",engine="python")
print(loans2007.iloc[0])
print(loans2007.shape[1])#注意shape[m,n],m是行，n是列

#特征越多越容易过拟合，因为可能某些特征价值很低，会给数据以噪声干扰
#去掉贷款后的信息，如funded_amnt/funded_amnt_inv;去掉重复的、高度相关的信息，如grade,sub_grade;去掉无法评估的信息，如emp_title
loans2007 = loans2007.drop(["id", "member_id", "funded_amnt", "funded_amnt_inv", "grade", "sub_grade", "emp_title", "issue_d"], axis=1)
loans2007 = loans2007.drop(["zip_code", "out_prncp", "out_prncp_inv", "total_pymnt", "total_pymnt_inv", "total_rec_prncp"], axis=1)
loans2007 = loans2007.drop(["total_rec_int", "total_rec_late_fee", "recoveries", "collection_recovery_fee", "last_pymnt_d", "last_pymnt_amnt"], axis=1)

#建立label值
#用value_counts()统计每一列的候选属性的出现次数
#取出'loan_status'属性值为 "Fully Paid"和"Charged Off"的行
#把字符值转化诶数值形式，用pandas.replace(),{选择的列{替代的键值对}}
print(loans2007["loan_status"].value_counts())
loans2007=loans2007[(loans2007["loan_status"]=="Fully Paid" )|( loans2007["loan_status"]=="Charged Off")]
status_replace={
    "
    ":{
        "Fully Paid":1,
        "Charged Off":0
    }
}
loans2007.replace(status_replace)

#去除 属性只有一个值 的列，但需要先去除nan缺失值
columns=loans2007.columns
drop_columns=[]
for col in columns:
    col_num=loans2007[col].dropna().unique()
    if len(col_num)== 1:
        drop_columns.append(col)
loans2007=loans2007.drop(drop_columns,axis=1)
print(drop_columns)
print(loans2007.shape)
loans2007.to_csv(r"E:\BaiduNetdiskDownload\唐宇迪机器学习\贷款利润最大化\filtered_loans2007.csv")

三、数据预处理

#读入数据
#统计缺失值
import pandas as pd
loans=pd.read_csv(r"E:\BaiduNetdiskDownload\唐宇迪机器学习\贷款利润最大化\filtered_loans2007.csv",engine="python")
null_counts=loans.isnull().sum()
print(null_counts)
print(loans["emp_length"].isnull().sum())   #emp_length的值为1073，为什么

#去掉"pub_rec_bankruptcies"特征列
#去掉有缺失值的行
#统计各个数据类型及其数量
#loans = loans.drop("pub_rec_bankruptcies", axis=1)
loans = loans.dropna(axis=0)
print(loans.dtypes.value_counts())

#select_dtypes选择类型为object的列，并查看第一行
object_columns=loans.select_dtypes(include=["object"])
print(object_columns.iloc[0])

#purpose和title的含义类似，可以只选其一
print(loans["purpose"].value_counts())
print(loans["title"].value_counts())

#对特征进行处理，可以和刚才一样，采用字典形式
#"去掉last_credit_pull_d,earliest_cr_line,addr_state,title列
#"int_tate","revol_util"的百分号去掉，之后转化为float
print(loans["emp_length"].unique())
mapping_dict={
    "emp_length":{
        '10+ years':10,
        '< 1 year':0,
        '3 years':3,
        '8 years' :8,
        '9 years' :9,
        '4 years' :4,
        '5 years':5,
        '1 year' :1,
        '6 years':6,
        '2 years' :2,
        '7 years':7
    }
}
loans=loans.drop(["last_credit_pull_d","earliest_cr_line","addr_state","title"],axis=1)
loans["int_rate"]=loans["int_rate"].str.rstrip("%").astype("float")
loans["revol_util"]=loans["revol_util"].str.rstrip("%").astype("float")
loans=loans.replace(mapping_dict)

cat_columns = ["home_ownership", "verification_status", "emp_length", "purpose", "term"]
dummy_df = pd.get_dummies(loans[cat_columns])#实现one hot encode
loans = pd.concat([loans, dummy_df], axis=1)#拼接到主体数据上
loans = loans.drop(cat_columns, axis=1) #丢除操作后的原数据
loans = loans.drop("pymnt_plan", axis=1)

loans.to_csv('cleaned_loans2007.csv', index=False)

四、获得最大利润的条件与做法

#读入数据
#info()样本个数，有无缺失，类型
import pandas as pd
loans=pd.read_csv("cleaned_loans2007.csv")
print(loans.info())

#准备train和test,label准备好之后开始训练模型
cols=loans.columns
train_col=cols.drop("loan_status")
features=loans[train_col]
target=loans["loan_status"]

#使用逻辑回归查看分类效果
#衡量tp,fp指标值，期望to高，fp低
#怀疑所有人都借了，打印前20条查看
from sklearn.linear_model import LinearRegression
from sklearn.model_selection import cross_val_predict,KFold
lr=LinearRegression()
kf=KFold(5,shuffle=False,random_state=1)
#交叉验证
predictions=cross_val_predict(lr,features,target,cv=kf)
predictions=pd.Series(predictions)
（未完待续）