DataScience：基于GiveMeSomeCredit数据集利用特征工程处理、逻辑回归LoR算法实现构建风控中的金融评分卡模型实现

基于GiveMeSomeCredit数据集利用特征工程处理、逻辑回归LoR算法实现构建风控中的金融评分卡模型

1、加载数据集

查看数据集的摘要信息

1.    Unnamed: 0  ...  NumberOfDependents
2. 0           1  ...                 2.0
3. 1           2  ...                 1.0
4. 2           3  ...                 0.0
5. 3           4  ...                 0.0
6. 4           5  ...                 0.0
7. 
8. [5 rows x 12 columns]
9. <class 'pandas.core.frame.DataFrame'>
10. RangeIndex: 150000 entries, 0 to 149999
11. Data columns (total 12 columns):
12.  #   Column                                Non-Null Count   Dtype  
13. ---  ------                                --------------   -----  
14. 0   Unnamed: 0                            150000 non-null  int64
15. 1   SeriousDlqin2yrs                      150000 non-null  int64
16. 2   RevolvingUtilizationOfUnsecuredLines  150000 non-null  float64
17. 3   age                                   150000 non-null  int64
18. 4   NumberOfTime30-59DaysPastDueNotWorse  150000 non-null  int64
19. 5   DebtRatio                             150000 non-null  float64
20. 6   MonthlyIncome                         120269 non-null  float64
21. 7   NumberOfOpenCreditLinesAndLoans       150000 non-null  int64
22. 8   NumberOfTimes90DaysLate               150000 non-null  int64
23. 9   NumberRealEstateLoansOrLines          150000 non-null  int64
24. 10  NumberOfTime60-89DaysPastDueNotWorse  150000 non-null  int64
25. 11  NumberOfDependents                    146076 non-null  float64
26. dtypes: float64(4), int64(8)
27. memory usage: 13.7 MB
28. None
29.           Unnamed: 0  ...  NumberOfDependents
30. count  150000.000000  ...       146076.000000
31. mean    75000.500000  ...            0.757222
32. std     43301.414527  ...            1.115086
33. min         1.000000  ...            0.000000
34. 25%     37500.750000  ...            0.000000
35. 50%     75000.500000  ...            0.000000
36. 75%    112500.250000  ...            1.000000
37. max    150000.000000  ...           20.000000

2、特征工程：数据分析与处理

# 2.1、缺失值分析及处理

1. [8 rows x 12 columns]
2. Column  Number_of_Null_Values  Proportion
3. 0                             Unnamed: 0                      0    0.000000
4. 1                       SeriousDlqin2yrs                      0    0.000000
5. 2   RevolvingUtilizationOfUnsecuredLines                      0    0.000000
6. 3                                    age                      0    0.000000
7. 4   NumberOfTime30-59DaysPastDueNotWorse                      0    0.000000
8. 5                              DebtRatio                      0    0.000000
9. 6                          MonthlyIncome                  29731    0.198207
10. 7        NumberOfOpenCreditLinesAndLoans                      0    0.000000
11. 8                NumberOfTimes90DaysLate                      0    0.000000
12. 9           NumberRealEstateLoansOrLines                      0    0.000000
13. 10  NumberOfTime60-89DaysPastDueNotWorse                      0    0.000000
14. 11                    NumberOfDependents                   3924    0.026160
15. Unnamed: 0                              0
16. SeriousDlqin2yrs                        0
17. RevolvingUtilizationOfUnsecuredLines    0
18. age                                     0
19. NumberOfTime30-59DaysPastDueNotWorse    0
20. DebtRatio                               0
21. MonthlyIncome                           0
22. NumberOfOpenCreditLinesAndLoans         0
23. NumberOfTimes90DaysLate                 0
24. NumberRealEstateLoansOrLines            0
25. NumberOfTime60-89DaysPastDueNotWorse    0
26. NumberOfDependents                      0

# 2.2、单个字段逐个分析

分析label字段：统计SeriousDlqin2yrs类别及其个数统计

1. Default Rate: 0.06684
2. count    150000.000000
3. mean          6.048438
4. std         249.755371
5. min           0.000000
6. 25%           0.029867
7. 50%           0.154181
8. 75%           0.559046
9. max       50708.000000
10. Name: RevolvingUtilizationOfUnsecuredLines, dtype: float64

1. [[0, 0.06684], [1, 0.37177950868783705], [2, 0.14555256064690028], [3, 0.09931506849315068], [4, 0.08679245283018867], [5, 0.07874015748031496], [6, 0.07692307692307693], [7, 0.0778688524590164], [8, 0.07407407407407407], [9, 0.07053941908713693], [10, 0.07053941908713693], [11, 0.07053941908713693], [12, 0.06666666666666667], [13, 0.058823529411764705], [14, 0.058823529411764705], [15, 0.05531914893617021], [16, 0.05531914893617021], [17, 0.05531914893617021], [18, 0.05531914893617021], [19, 0.05555555555555555]]
2. Proportion of Defaulters with Total Amount of Money Owed Not Exceeding Total Credit Limit: 0.05991996127598361
3. Proportion of Defaulters with Total Amount of Money Owed Not Exceeding or Equal to 13 times of Total Credit Limit:
4. 0.06685273968029273

分析age字段

1. count    150000.000000
2. mean         52.295207
3. std          14.771866
4. min           0.000000
5. 25%          41.000000
6. 50%          52.000000
7. 75%          63.000000
8. max         109.000000
9. Name: age, dtype: float64

分析3个类似字段—NumberOfTimes90DaysLate、NumberOfTime60

89DaysPastDueNotWorse、NumberOfTime30-59DaysPastDueNotWorse

1. 0     141662
2. 1       5243
3. 2       1555
4. 3        667
5. 4        291
6. 5        131
7. 6         80
8. 7         38
9. 8         21
10. 9         19
11. 10         8
12. 11         5
13. 12         2
14. 13         4
15. 14         2
16. 15         2
17. 17         1
18. 96         5
19. 98       264
20. Name: NumberOfTimes90DaysLate, dtype: int64 
21. 0     142396
22. 1       5731
23. 2       1118
24. 3        318
25. 4        105
26. 5         34
27. 6         16
28. 7          9
29. 8          2
30. 9          1
31. 11         1
32. 96         5
33. 98       264
34. Name: NumberOfTime60-89DaysPastDueNotWorse, dtype: int64 
35. 0     126018
36. 1      16033
37. 2       4598
38. 3       1754
39. 4        747
40. 5        342
41. 6        140
42. 7         54
43. 8         25
44. 9         12
45. 10         4
46. 11         1
47. 12         2
48. 13         1
49. 96         5
50. 98       264
51. Name: NumberOfTime30-59DaysPastDueNotWorse, dtype: int64
52.        NumberOfTimes90DaysLate  ...  NumberOfTime30-59DaysPastDueNotWorse
53. count               269.000000  ...                            269.000000
54. mean                 97.962825  ...                             97.962825
55. std                   0.270628  ...                              0.270628
56. min                  96.000000  ...                             96.000000
57. 25%                  98.000000  ...                             98.000000
58. 50%                  98.000000  ...                             98.000000
59. 75%                  98.000000  ...                             98.000000
60. max                  98.000000  ...                             98.000000
61. 
62. [8 rows x 3 columns]
63. {'98,98,98': 263, '96,96,96': 4}

分析单个字段—DebtRatio及与MonthlyIncome、SeriousDlqin2yrs关系

1.     temp = df_train[(df_DR > df_DR95) & (df_train['SeriousDlqin2yrs'] == df_train['MonthlyIncome'])]
2.     temp.to_csv('20220314temp.csv')

1. count    150000.000000
2. mean        353.005076
3. std        2037.818523
4. min           0.000000
5. 25%           0.175074
6. 50%           0.366508
7. 75%           0.868254
8. max      329664.000000
9. Name: DebtRatio, dtype: float64 
10. 2449.0
11.             DebtRatio  MonthlyIncome  SeriousDlqin2yrs
12. count    7494.000000    7494.000000       7494.000000
13. mean     4417.958367    5126.905791          0.055111
14. std      7875.314649    1183.339377          0.228212
15. min      2450.000000       0.000000          0.000000
16. 25%      2893.250000    5400.000000          0.000000
17. 50%      3491.000000    5400.000000          0.000000
18. 75%      4620.000000    5400.000000          0.000000
19. max    329664.000000    5400.000000          1.000000
20. 331
21. 5400.0    7115
22. 0.0        347
23. 1.0         32
24. Name: MonthlyIncome, dtype: int64
25. Number of people who owe around 2449 or more times what they own and have same values for MonthlyIncome and SeriousDlqin2yrs: 331
26. 3489.024999999994
27.             DebtRatio  MonthlyIncome  SeriousDlqin2yrs
28. count    3750.000000     3750.00000       3750.000000
29. mean     5917.488000     5133.60320          0.064267
30. std     10925.524011     1169.58239          0.245260
31. min      3490.000000        0.00000          0.000000
32. 25%      3957.250000     5400.00000          0.000000
33. 50%      4619.000000     5400.00000          0.000000
34. 75%      5789.500000     5400.00000          0.000000
35. max    329664.000000     5400.00000          1.000000
36. 164
37. 5400.0    3565
38. 0.0        173
39. 1.0         12
40. Name: MonthlyIncome, dtype: int64
41. Number of people who owe around 3490 or more times what they own and have same values for MonthlyIncome and SeriousDlqin2yrs: 164

分析单个字段—MonthlyIncome

分析单个字段—NumberOfOpenCreditLinesAndLoans

分析单个字段—NumberRealEstateLoansOrLines

分析单个字段—NumberOfDependents

# 2.3、数据分箱

仅label没分箱处理

# 2.4、特征筛选：利用IV方法

1. bin_DebtRatio cal_IV:  0.0595
2. bin_MonthlyIncome cal_IV:  0.0562
3. bin_RevolvingUtilizationOfUnsecuredLines cal_IV:  1.0596
4. bin_NumberOfOpenCreditLinesAndLoans cal_IV:  0.048
5. bin_NumberRealEstateLoansOrLines cal_IV:  0.0121
6. bin_age cal_IV:  0.2404
7. bin_NumberOfDependents cal_IV:  0.0145
8. bin_NumberOfTime30-59DaysPastDueNotWorse cal_IV:  0.4924
9. bin_NumberOfTime60-89DaysPastDueNotWorse cal_IV:  0.2666
10. bin_NumberOfTimes90DaysLate cal_IV:  0.4916

# 2.5、计算WOE值

# 2.5.1、基于筛选的特征，利用WOE函数把分箱转成WOE值

woe_cols： ['woe_bin_age', 'woe_bin_RevolvingUtilizationOfUnsecuredLines', 'woe_bin_NumberOfTime30-59DaysPastDueNotWorse', 'woe_bin_NumberOfTime60-89DaysPastDueNotWorse', 'woe_bin_NumberOfTimes90DaysLate']

1. ------------- age
2. <class 'pandas.core.frame.DataFrame'> df……final 
3.     features           bin       woe
4. 0       age  (40.0, 50.0]  0.228343
5. 1       age  (25.0, 40.0]  0.469547
6. 5       age   (70.0, inf] -1.132145
7. 6       age  (50.0, 60.0] -0.084782
8. 15      age  (60.0, 70.0] -0.689003
9. 19      age  (-inf, 25.0]  0.562024
10. 
11. 
12. ------------- RevolvingUtilizationOfUnsecuredLines
13. <class 'pandas.core.frame.DataFrame'> df……final 
14.                                  features               bin       woe
15. 0   RevolvingUtilizationOfUnsecuredLines  (0.699, 50708.0]  1.242254
16. 2   RevolvingUtilizationOfUnsecuredLines    (0.271, 0.699]  0.053164
17. 3   RevolvingUtilizationOfUnsecuredLines   (0.0832, 0.271] -0.866502
18. 11  RevolvingUtilizationOfUnsecuredLines  (-0.001, 0.0192] -1.286617
19. 14  RevolvingUtilizationOfUnsecuredLines  (0.0192, 0.0832] -1.447382
20. 
21. 
22. ------------- NumberOfTime30-59DaysPastDueNotWorse
23. <class 'pandas.core.frame.DataFrame'> df……final 
24.                                     features          bin       woe
25. 0      NumberOfTime30-59DaysPastDueNotWorse   (1.0, 2.0]  1.616726
26. 1      NumberOfTime30-59DaysPastDueNotWorse  (-inf, 1.0] -0.257826
27. 13     NumberOfTime30-59DaysPastDueNotWorse   (2.0, 3.0]  2.027495
28. 183    NumberOfTime30-59DaysPastDueNotWorse   (3.0, 4.0]  2.336869
29. 191    NumberOfTime30-59DaysPastDueNotWorse   (4.0, 5.0]  2.436786
30. 251    NumberOfTime30-59DaysPastDueNotWorse   (6.0, 7.0]  2.710383
31. 423    NumberOfTime30-59DaysPastDueNotWorse   (9.0, inf]  2.846431
32. 1052   NumberOfTime30-59DaysPastDueNotWorse   (5.0, 6.0]  2.750685
33. 6909   NumberOfTime30-59DaysPastDueNotWorse   (7.0, 8.0]  1.882503
34. 10822  NumberOfTime30-59DaysPastDueNotWorse   (8.0, 9.0]  1.943128
35. 
36. 
37. ------------- NumberOfTime60-89DaysPastDueNotWorse
38. <class 'pandas.core.frame.DataFrame'> df……final 
39.                                     features          bin       woe
40. 0      NumberOfTime60-89DaysPastDueNotWorse  (-inf, 1.0] -0.097990
41. 186    NumberOfTime60-89DaysPastDueNotWorse   (1.0, 2.0]  2.643431
42. 423    NumberOfTime60-89DaysPastDueNotWorse   (4.0, 5.0]  3.115848
43. 1146   NumberOfTime60-89DaysPastDueNotWorse   (2.0, 3.0]  2.901978
44. 1733   NumberOfTime60-89DaysPastDueNotWorse   (9.0, inf]  2.829466
45. 2406   NumberOfTime60-89DaysPastDueNotWorse   (3.0, 4.0]  3.121783
46. 6664   NumberOfTime60-89DaysPastDueNotWorse   (5.0, 6.0]  3.734887
47. 16642  NumberOfTime60-89DaysPastDueNotWorse   (6.0, 7.0]  2.859419
48. 23964  NumberOfTime60-89DaysPastDueNotWorse   (7.0, 8.0]  2.636275
49. 68976  NumberOfTime60-89DaysPastDueNotWorse   (8.0, 9.0]  2.636275
50. 
51. 
52. ------------- NumberOfTimes90DaysLate
53. <class 'pandas.core.frame.DataFrame'> df……final 
54.                       features          bin       woe
55. 0     NumberOfTimes90DaysLate  (-inf, 1.0] -0.176674
56. 13    NumberOfTimes90DaysLate   (2.0, 3.0]  2.947611
57. 186   NumberOfTimes90DaysLate   (1.0, 2.0]  2.632416
58. 1298  NumberOfTimes90DaysLate   (4.0, 5.0]  3.183915
59. 1713  NumberOfTimes90DaysLate   (3.0, 4.0]  3.344926
60. 1733  NumberOfTimes90DaysLate   (9.0, inf]  2.821100
61. 2910  NumberOfTimes90DaysLate   (8.0, 9.0]  3.665894
62. 3400  NumberOfTimes90DaysLate   (5.0, 6.0]  3.041740
63. 3929  NumberOfTimes90DaysLate   (6.0, 7.0]  4.124352
64. 5684  NumberOfTimes90DaysLate   (7.0, 8.0]  3.552566

# 2.5.2、解析不同bin对应woe值的一一对应情况

Variable	Binning	Score
NumberOfTime30-59DaysPastDueNotWorse	(-inf, 1.0]	11
NumberOfTime30-59DaysPastDueNotWorse	(1.0, 2.0]	-70
NumberOfTime30-59DaysPastDueNotWorse	(2.0, 3.0]	-87
NumberOfTime30-59DaysPastDueNotWorse	(3.0, 4.0]	-101
NumberOfTime30-59DaysPastDueNotWorse	(4.0, 5.0]	-105
NumberOfTime30-59DaysPastDueNotWorse	(5.0, 6.0]	-119
NumberOfTime30-59DaysPastDueNotWorse	(6.0, 7.0]	-117
NumberOfTime30-59DaysPastDueNotWorse	(7.0, 8.0]	-81
NumberOfTime30-59DaysPastDueNotWorse	(8.0, 9.0]	-84
NumberOfTime30-59DaysPastDueNotWorse	(9.0, inf]	-123

NumberOfTime60-89DaysPastDueNotWorse	(-inf, 1.0]	2
NumberOfTime60-89DaysPastDueNotWorse	(1.0, 2.0]	-66
NumberOfTime60-89DaysPastDueNotWorse	(2.0, 3.0]	-73
NumberOfTime60-89DaysPastDueNotWorse	(3.0, 4.0]	-78
NumberOfTime60-89DaysPastDueNotWorse	(4.0, 5.0]	-78
NumberOfTime60-89DaysPastDueNotWorse	(5.0, 6.0]	-94
NumberOfTime60-89DaysPastDueNotWorse	(6.0, 7.0]	-72
NumberOfTime60-89DaysPastDueNotWorse	(7.0, 8.0]	-66
NumberOfTime60-89DaysPastDueNotWorse	(8.0, 9.0]	-66
NumberOfTime60-89DaysPastDueNotWorse	(9.0, inf]	-71
NumberOfTimes90DaysLate	(-inf, 1.0]	7
NumberOfTimes90DaysLate	(1.0, 2.0]	-107
NumberOfTimes90DaysLate	(2.0, 3.0]	-120
NumberOfTimes90DaysLate	(3.0, 4.0]	-137
NumberOfTimes90DaysLate	(4.0, 5.0]	-130
NumberOfTimes90DaysLate	(5.0, 6.0]	-124
NumberOfTimes90DaysLate	(6.0, 7.0]	-168
NumberOfTimes90DaysLate	(7.0, 8.0]	-145
NumberOfTimes90DaysLate	(8.0, 9.0]	-150
NumberOfTimes90DaysLate	(9.0, inf]	-115
RevolvingUtilizationOfUnsecuredLines	(-0.001, 0.0192]	71
RevolvingUtilizationOfUnsecuredLines	(0.0192, 0.0832]	80
RevolvingUtilizationOfUnsecuredLines	(0.0832, 0.271]	48
RevolvingUtilizationOfUnsecuredLines	(0.271, 0.699]	-3
RevolvingUtilizationOfUnsecuredLines	(0.699, 50708.0]	-69
age	(-inf, 25.0]	-19
age	(25.0, 40.0]	-16
age	(40.0, 50.0]	-8
age	(50.0, 60.0]	3
age	(60.0, 70.0]	23
age	(70.0, inf]	38

# 2.6、切分数据集：留25%作为模型的验证集

1. bad_rate:  0.06688333333333334
2. X_train.shape: (120000, 5)

# 3、逻辑回归建模

# 3.1、建立模型

1. LoR_Score： 0.9368266666666667
2. LoRC_pred_proba [0.0121424  0.15221691 0.02248172 ... 0.0528182  0.0121424  0.0952767 ]
3. LoRC_coef_lists_ 
4.  [0.46051155 0.76869053 0.59104431 0.36452944 0.56621256]

# 3.2、模型评估：计算AUC值、绘制ROC曲线、输出混淆矩阵

1. Auc_Score： 0.8226466762033763
2. [[34827   200]
3.  [ 2169   304]]

# 4、模型推理

# 4.1、设计评分卡规则表

# 4.1.1、求出两个刻度A、B：根据2个假设推导出评分卡的刻度参数A和B计算公式

650 72.13

# 4.1.2、设计评分卡规则表：根据刻度B、对应分箱的WOE编码、模型系数，得到score_card_rule

	Variable	Binning	Score
0	age	(40.0, 50.0]	-8
1	age	(25.0, 40.0]	-16
2	age	(70.0, inf]	38
3	age	(50.0, 60.0]	3
4	age	(60.0, 70.0]	23
5	age	(-inf, 25.0]	-19
6	RevolvingUtilizationOfUnsecuredLines	(0.699, 50708.0]	-69
7	RevolvingUtilizationOfUnsecuredLines	(0.271, 0.699]	-3
8	RevolvingUtilizationOfUnsecuredLines	(0.0832, 0.271]	48
9	RevolvingUtilizationOfUnsecuredLines	(-0.001, 0.0192]	71
10	RevolvingUtilizationOfUnsecuredLines	(0.0192, 0.0832]	80
11	NumberOfTime30-59DaysPastDueNotWorse	(1.0, 2.0]	-69
12	NumberOfTime30-59DaysPastDueNotWorse	(-inf, 1.0]	11
13	NumberOfTime30-59DaysPastDueNotWorse	(2.0, 3.0]	-86
14	NumberOfTime30-59DaysPastDueNotWorse	(3.0, 4.0]	-100
15	NumberOfTime30-59DaysPastDueNotWorse	(4.0, 5.0]	-104
16	NumberOfTime30-59DaysPastDueNotWorse	(6.0, 7.0]	-116
17	NumberOfTime30-59DaysPastDueNotWorse	(9.0, inf]	-121
18	NumberOfTime30-59DaysPastDueNotWorse	(5.0, 6.0]	-117
19	NumberOfTime30-59DaysPastDueNotWorse	(7.0, 8.0]	-80
20	NumberOfTime30-59DaysPastDueNotWorse	(8.0, 9.0]	-83
21	NumberOfTime60-89DaysPastDueNotWorse	(-inf, 1.0]	3
22	NumberOfTime60-89DaysPastDueNotWorse	(1.0, 2.0]	-70
23	NumberOfTime60-89DaysPastDueNotWorse	(4.0, 5.0]	-82
24	NumberOfTime60-89DaysPastDueNotWorse	(2.0, 3.0]	-76
25	NumberOfTime60-89DaysPastDueNotWorse	(9.0, inf]	-74
26	NumberOfTime60-89DaysPastDueNotWorse	(3.0, 4.0]	-82
27	NumberOfTime60-89DaysPastDueNotWorse	(5.0, 6.0]	-98
28	NumberOfTime60-89DaysPastDueNotWorse	(6.0, 7.0]	-75
29	NumberOfTime60-89DaysPastDueNotWorse	(7.0, 8.0]	-69
30	NumberOfTime60-89DaysPastDueNotWorse	(8.0, 9.0]	-69
31	NumberOfTimes90DaysLate	(-inf, 1.0]	7
32	NumberOfTimes90DaysLate	(2.0, 3.0]	-120
33	NumberOfTimes90DaysLate	(1.0, 2.0]	-108
34	NumberOfTimes90DaysLate	(4.0, 5.0]	-130
35	NumberOfTimes90DaysLate	(3.0, 4.0]	-137
36	NumberOfTimes90DaysLate	(9.0, inf]	-115
37	NumberOfTimes90DaysLate	(8.0, 9.0]	-150
38	NumberOfTimes90DaysLate	(5.0, 6.0]	-124
39	NumberOfTimes90DaysLate	(6.0, 7.0]	-168
40	NumberOfTimes90DaysLate	(7.0, 8.0]	-145

# 4.2、结合刻度A计算样本评分卡得分

# 4.2.1、随机选取12个样本(6个好的和6个坏的)并计算每个样本的总评分并对比Label，可验证模型效果

# 4.2.2、结合刻度A计算样本评分卡得分

	age	RevolvingUtilization OfUnsecuredLines	NumberOfTime 30-59Days PastDueNotWorse	NumberOfTime60-89 DaysPastDueNotWorse	NumberOf Times90 DaysLate	score
44377	55	0.081686933	0	0	0	754
25143	47	0.9999999	0	0	1	594
67429	54	0.015170898	0	0	0	745
66689	26	0.252252252	0	0	0	703
42656	40	0.916334661	1	0	0	586
81903	65	0.091477937	0	0	0	742

	age	Revolving UtilizationOf UnsecuredLines	NumberOfTime 30-59Days PastDueNotWorse	NumberOfTime60-89 DaysPastDueNotWorse	NumberOf Times90 DaysLate	score
111052	30	0.9999999	0	4	2	386
30582	30	0.9999999	0	0	0	586
23677	43	0.68756082	0	0	0	660
87669	27	0.9999999	0	1	1	586
46920	50	0.442370466	0	0	0	660
78952	48	0.40781316	0	0	0	660

# 4.3、对比测试样本得分及其对应标签，进而设计评审策略

1. 44377 754.0 --------- 直接接受！
2. age                                      47.0
3. RevolvingUtilizationOfUnsecuredLines      1.0
4. NumberOfTime30-59DaysPastDueNotWorse      0.0
5. NumberOfTime60-89DaysPastDueNotWorse      0.0
6. NumberOfTimes90DaysLate                   1.0
7. score                                   594.0
8. Name: 25143, dtype: float64
9. 25143 594.0 --------- 人工审核！
10. age                                      54.000000
11. RevolvingUtilizationOfUnsecuredLines      0.015171
12. NumberOfTime30-59DaysPastDueNotWorse      0.000000
13. NumberOfTime60-89DaysPastDueNotWorse      0.000000
14. NumberOfTimes90DaysLate                   0.000000
15. score                                   745.000000
16. Name: 67429, dtype: float64
17. 67429 745.0 --------- 直接接受！
18. age                                      26.000000
19. RevolvingUtilizationOfUnsecuredLines      0.252252
20. NumberOfTime30-59DaysPastDueNotWorse      0.000000
21. NumberOfTime60-89DaysPastDueNotWorse      0.000000
22. NumberOfTimes90DaysLate                   0.000000
23. score                                   703.000000
24. Name: 66689, dtype: float64
25. 66689 703.0 --------- 直接接受！
26. age                                      40.000000
27. RevolvingUtilizationOfUnsecuredLines      0.916335
28. NumberOfTime30-59DaysPastDueNotWorse      1.000000
29. NumberOfTime60-89DaysPastDueNotWorse      0.000000
30. NumberOfTimes90DaysLate                   0.000000
31. score                                   586.000000
32. Name: 42656, dtype: float64
33. 42656 586.0 --------- 人工审核！
34. age                                      65.000000
35. RevolvingUtilizationOfUnsecuredLines      0.091478
36. NumberOfTime30-59DaysPastDueNotWorse      0.000000
37. NumberOfTime60-89DaysPastDueNotWorse      0.000000
38. NumberOfTimes90DaysLate                   0.000000
39. score                                   742.000000
40. Name: 81903, dtype: float64
41. 81903 742.0 --------- 直接接受！

DataScience：基于GiveMeSomeCredit数据集利用特征工程处理、逻辑回归LoR算法实现构建风控中的金融评分卡模型

基于GiveMeSomeCredit数据集利用特征工程处理、逻辑回归LoR算法实现构建风控中的金融评分卡模型

1、加载数据集

查看数据集的摘要信息

2、特征工程：数据分析与处理

# 2.1、缺失值分析及处理

# 2.2、单个字段逐个分析

分析label字段：统计SeriousDlqin2yrs类别及其个数统计

分析age字段

分析3个类似字段—NumberOfTimes90DaysLate、NumberOfTime60

分析单个字段—DebtRatio及与MonthlyIncome、SeriousDlqin2yrs关系

分析单个字段—MonthlyIncome

分析单个字段—NumberOfOpenCreditLinesAndLoans

分析单个字段—NumberRealEstateLoansOrLines

分析单个字段—NumberOfDependents

# 2.3、数据分箱

# 2.4、特征筛选：利用IV方法

# 2.5、计算WOE值

# 2.5.1、基于筛选的特征，利用WOE函数把分箱转成WOE值

# 2.5.2、解析不同bin对应woe值的一一对应情况

# 2.6、切分数据集：留25%作为模型的验证集

# 3、逻辑回归建模

# 3.1、建立模型

# 3.2、模型评估：计算AUC值、绘制ROC曲线、输出混淆矩阵

# 4、模型推理

# 4.1、设计评分卡规则表

# 4.1.1、求出两个刻度A、B：根据2个假设推导出评分卡的刻度参数A和B计算公式

# 4.1.2、设计评分卡规则表：根据刻度B、对应分箱的WOE编码、模型系数，得到score_card_rule

# 4.2、结合刻度A计算样本评分卡得分

# 4.2.1、随机选取12个样本(6个好的和6个坏的)并计算每个样本的总评分并对比Label，可验证模型效果

# 4.2.2、结合刻度A计算样本评分卡得分

# 4.3、对比测试样本得分及其对应标签，进而设计评审策略

热门文章

最新文章

相关课程

相关电子书

相关实验场景

热门

活动广场

任务中心

开发者评测

高校计划

乘风者计划

训练营

阿里云MVP

话题

直播

下载

镜像站

技术资料

插件

DataScience：基于GiveMeSomeCredit数据集利用特征工程处理、逻辑回归LoR算法实现构建风控中的金融评分卡模型

基于GiveMeSomeCredit数据集利用特征工程处理、逻辑回归LoR算法实现构建风控中的金融评分卡模型

1、加载数据集

查看数据集的摘要信息

2、特征工程：数据分析与处理

# 2.1、缺失值分析及处理

# 2.2、单个字段逐个分析

分析label字段：统计SeriousDlqin2yrs类别及其个数统计

分析age字段

分析3个类似字段—NumberOfTimes90DaysLate、NumberOfTime60

分析单个字段—DebtRatio及与MonthlyIncome、SeriousDlqin2yrs关系

分析单个字段—MonthlyIncome

分析单个字段—NumberOfOpenCreditLinesAndLoans

分析单个字段—NumberRealEstateLoansOrLines

分析单个字段—NumberOfDependents

# 2.3、数据分箱

# 2.4、特征筛选：利用IV方法

# 2.5、计算WOE值

# 2.5.1、基于筛选的特征，利用WOE函数把分箱转成WOE值

# 2.5.2、解析不同bin对应woe值的一一对应情况

# 2.6、切分数据集：留25%作为模型的验证集

# 3、逻辑回归建模

# 3.1、建立模型

# 3.2、模型评估：计算AUC值、绘制ROC曲线、输出混淆矩阵

# 4、模型推理

# 4.1、设计评分卡规则表

# 4.1.1、求出两个刻度A、B：根据2个假设推导出评分卡的刻度参数A和B计算公式

# 4.1.2、设计评分卡规则表 ：根据刻度B、对应分箱的WOE编码、模型系数，得到score_card_rule

# 4.2、结合刻度A计算样本评分卡得分

# 4.2.1、随机选取12个样本(6个好的和6个坏的)并计算每个样本的总评分并对比Label，可验证模型效果

# 4.2.2、结合刻度A计算样本评分卡得分

# 4.3、对比测试样本得分及其对应标签，进而设计评审策略

热门文章

最新文章

相关课程

相关电子书

相关实验场景

# 4.1.2、设计评分卡规则表：根据刻度B、对应分箱的WOE编码、模型系数，得到score_card_rule