04 分类算法 - Logistic回归 - 信贷审批案例

需求：

基于信贷数据进行用户信贷分类，使用Logistic算法构建模型，并比较这两大类算法的效果

数据来源：

http://archive.ics.uci.edu/ml/datasets/Credit+Approval

数据格式：

b,30.83,0,u,g,w,v,1.25,t,t,01,f,g,00202,0,+
a,58.67,4.46,u,g,q,h,3.04,t,t,06,f,g,00043,560,+
a,24.50,0.5,u,g,q,h,1.5,t,f,0,f,g,00280,824,+
b,27.83,1.54,u,g,w,v,3.75,t,t,05,t,g,00100,3,+
b,20.17,5.625,u,g,w,v,1.71,t,f,0,f,s,00120,0,+
b,32.08,4,u,g,m,v,2.5,t,f,0,t,g,00360,0,+
b,33.17,1.04,u,g,r,h,6.5,t,f,0,t,g,00164,31285,+
a,22.92,11.585,u,g,cc,v,0.04,t,f,0,f,g,00080,1349,+

import numpy as np
import matplotlib as mpl
import matplotlib.pyplot as plt
import pandas as pd
import warnings
import sklearn
from sklearn.linear_model import LogisticRegressionCV
from sklearn.linear_model.coordinate_descent import ConvergenceWarning
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import StandardScaler
from sklearn.pipeline import Pipeline
from sklearn.neighbors import KNeighborsClassifier

## 设置字符集，防止中文乱码
mpl.rcParams['font.sans-serif']=[u'simHei']
mpl.rcParams['axes.unicode_minus']=False
## 拦截异常
warnings.filterwarnings(action = 'ignore', category=ConvergenceWarning)

加载数据并对数据进行预处理:

# 1. 加载数据
path = "datas/crx.data"
names = ['A1','A2','A3','A4','A5','A6','A7','A8',
         'A9','A10','A11','A12','A13','A14','A15','A16']
df = pd.read_csv(path, header=None, names=names)
print ("数据条数:", len(df))

# 2. 异常数据过滤
df = df.replace("?", np.nan).dropna(how='any')
print ("过滤后数据条数:", len(df))

df.head(5)

数据条数: 690
过滤后数据条数: 653

了解A6特征的内容：

df['A6'].value_counts()

c 133
q 75
w 63
i 55
aa 52
ff 50
k 48
cc 40
m 38
x 36
d 26
e 24
j 10
r 3
Name: A6, dtype: int64

对部分特征进行哑编码操作:

def parse(v, l):
    return [1 if i == v else 0 for i in l]
def parseRecord(record):
    result = []
    ## 格式化数据，将离线数据转换为连续数据
    a1 = record['A1']
    for i in parse(a1, ('a', 'b')):
        result.append(i)
    
    result.append(float(record['A2']))
    result.append(float(record['A3']))
    
    
    a4 = record['A4']
    for i in parse(a4, ('u', 'y', 'l', 't')):
        result.append(i)
    
    a5 = record['A5']
    for i in parse(a5, ('g', 'p', 'gg')):
        result.append(i)
    
    a6 = record['A6']
    for i in parse(a6, ('c', 'd', 'cc', 'i', 'j', 'k', 'm', 'r', 'q', 'w', 'x', 'e', 'aa', 'ff')):
        result.append(i)
    
    a7 = record['A7']
    for i in parse(a7, ('v', 'h', 'bb', 'j', 'n', 'z', 'dd', 'ff', 'o')):
        result.append(i)
    
    result.append(float(record['A8']))
    
    a9 = record['A9']
    for i in parse(a9, ('t', 'f')):
        result.append(i)
        
    a10 = record['A10']
    for i in parse(a10, ('t', 'f')):
        result.append(i)
    
    result.append(float(record['A11']))
    
    a12 = record['A12']
    for i in parse(a12, ('t', 'f')):
        result.append(i)
        
    a13 = record['A13']
    for i in parse(a13, ('g', 'p', 's')):
        result.append(i)
    
    result.append(float(record['A14']))
    result.append(float(record['A15']))
    
    a16 = record['A16']
    if a16 == '+':
        result.append(1)
    else:
        result.append(0)
        
    return result

数据特征处理(将数据转换为数值类型的):

new_names =  ['A1_0', 'A1_1',
              'A2','A3',
              'A4_0','A4_1','A4_2','A4_3',
              'A5_0', 'A5_1', 'A5_2', 
              'A6_0', 'A6_1', 'A6_2', 'A6_3', 'A6_4', 'A6_5', 'A6_6', 
                'A6_7', 'A6_8', 'A6_9', 'A6_10', 'A6_11',
                     'A6_12', 'A6_13', 
              'A7_0', 'A7_1', 'A7_2', 'A7_3', 'A7_4', 'A7_5', 'A7_6', 
                 'A7_7', 'A7_8', 
              'A8',
              'A9_0', 'A9_1' ,
              'A10_0', 'A10_1',
              'A11',
              'A12_0', 'A12_1',
              'A13_0', 'A13_1', 'A13_2',
              'A14','A15','A16']
datas = df.apply(lambda x: pd.Series(parseRecord(x), 
    index = new_names), axis=1)
names = new_names

## 展示一下处理后的数据
datas.head(5)

数据分割:

X = datas[names[0:-1]]
Y = datas[names[-1]]

X_train,X_test,Y_train,Y_test = 
  train_test_split(X,Y,test_size=0.1,random_state=0)

X_train.describe().T

Logistic算法模型构建(重点)：

lr = LogisticRegressionCV(Cs=np.logspace(-4,1,10), 
  fit_intercept=True, penalty='l2', solver='lbfgs', tol=0.01, 
    multi_class='ovr')
lr.fit(X_train, Y_train)

LogisticRegressionCV(Cs=array([1.00000e-04, 3.59381e-04, 1.29155e-03, 4.64159e-03, 1.66810e-02,
5.99484e-02, 2.15443e-01, 7.74264e-01, 2.78256e+00, 1.00000e+01]),
class_weight=None, cv=None, dual=False, fit_intercept=True,
intercept_scaling=1.0, max_iter=100, multi_class='ovr',
n_jobs=1, penalty='l2', random_state=None, refit=True,
scoring=None, solver='lbfgs', tol=0.01, verbose=0)

Logistic算法效果输出(重点):

lr_r = lr.score(X_train, Y_train)
print ("Logistic算法准确率：", lr_r)
print ("Logistic算法稀疏化特征比率：%.2f%%" 
  % (np.mean(lr.coef_.ravel() == 0) * 100))
print ("Logistic算法参数：",lr.coef_)
print ("Logistic算法截距：",lr.intercept_)

Logistic算法准确率： 0.889267461669506
Logistic算法稀疏化特征比率：2.13%
Logistic算法参数： [[-0.02376298 0.02376298 0.10977985 -0.05129678 0.0706074 -0.09335579
0.16299053 0. 0.0706074 -0.09335579 0.16299053 0.01032474
0.02940631 0.24367798 -0.24817577 -0.26062573 -0.15506122 -0.02849042
-0.09242933 0.00886997 0.12227728 0.50973846 0.25210634 -0.05958224
-0.36722359 0.01038447 0.11584139 -0.07895233 0.24565932 0.18756209
-0.29710075 -0.08548594 -0.09832244 -0.05788513 0.26194759 0.85940096
-0.85940096 0.14167783 -0.14167783 0.53642415 -0.06228778 0.06228778
0.0237657 -0.07649129 -0.0080863 -0.39691472 0.80897268]]
Logistic算法截距： [-0.40412228]

Logistic算法预测（预测所属类别）:

lr_y_predict = lr.predict(X_test)

Logistic算法获取概率值(就是Logistic算法计算出来的结果值):

y1 = lr.predict_proba(X_test)

array([[ 1.13982099e-01, 8.86017901e-01],
[ 6.25759412e-02, 9.37424059e-01],
[ 5.07692157e-02, 9.49230784e-01],
[ 9.19448211e-01, 8.05517889e-02],
[ 4.34968508e-01, 5.65031492e-01],
[ 9.45823923e-05, 9.99905418e-01],
[ 9.51970606e-01, 4.80293939e-02],
[ 2.07954778e-02, 9.79204522e-01],
....

结果图像展示:

x_len = range(len(X_test))
plt.figure(figsize=(14,7), facecolor='w')
plt.ylim(-0.1,1.1)
plt.plot(x_len, Y_test, 'ro',markersize = 6, zorder=3, label=u'真实值')
plt.plot(x_len, lr_y_predict, 'go', markersize = 10, zorder=2, label='Logis算法预测值,准确率=%.3f' % lr.score(X_test, Y_test))
plt.legend(loc = 'center right')
plt.xlabel('数据编号', fontsize=18)
plt.ylabel('是否审批(0表示通过，1表示通过)', fontsize=18)
plt.show()