风控模型实验报告

实验 ID: 20260611-135656-3129 · 生成时间: 2026-06-11 05:56 UTC

1. 数据审计

样本数1000
字段数24
坏样本率36.30%
高风险字段5

审计建议

  • 列 `customer_id` 唯一率过高(0.999),可能是 ID 列,应排除在特征之外
  • 列 `apply_time` 唯一率过高(1.0),可能是 ID 列,应排除在特征之外
  • 列 `income` 唯一率过高(0.999),可能是 ID 列,应排除在特征之外
  • 列 `loan_amount` 唯一率过高(0.998),可能是 ID 列,应排除在特征之外
  • 目标变量 `historical_overdue_count` 非二值,取值: [np.int64(0), np.int64(1), np.int64(2), np.int64(3), np.int64(4)]
  • ID 列 `customer_id` 有 1 个重复值

2. 样本拆分

拆分方式: 时间切分,总样本 1000,目标非空 1000

数据集样本数比例坏样本率时间范围
训练集60060.00%35.00%2024-01-01 ~ 2024-01-25
验证集20020.00%36.00%2024-01-26 ~ 2024-02-03
OOT 集20020.00%40.50%2024-02-03 ~ 2024-02-11

3. 变量筛选

候选变量 21,入模 19,IV 阈值 0.01

变量IV单调分箱数
gender0.79662
marital_status0.72424
education0.59344
channel0.57804
purpose0.49906
fpd300.45201
constant_flag0.45201
clicks_before_apply0.40182
time_on_page_sec0.40032
months_since_last_loan0.39452
loan_term0.37742
region0.36987
existing_loans0.36493
credit_score0.36312
app_sessions_30d0.35942
city_tier0.35782
employment_years0.34893
age0.26484
debt_ratio0.24126

4. 模型评估

模型数据集AUCKSLift@10%PSI
LR + WOE 评分卡训练集0.19251.12
LR + WOE 评分卡验证集0.13000.75
LR + WOE 评分卡OOT 集0.19211.300.1105

LR + WOE 评分卡 — 训练集 坏样本率分档

分组样本数坏样本坏样本率分数范围
1603761.67%0.28 ~ 0.31
2603660.00%0.31 ~ 0.33
3604981.67%0.33 ~ 0.33
4604371.67%0.33 ~ 0.34
5604371.67%0.34 ~ 0.34
6604371.67%0.34 ~ 0.35
7605185.00%0.35 ~ 0.36
8604066.67%0.36 ~ 0.37
9604778.33%0.37 ~ 0.39
10604981.67%0.39 ~ 0.48

LR + WOE 评分卡 — 验证集 坏样本率分档

分组样本数坏样本坏样本率分数范围
1201260.00%0.28 ~ 0.31
220735.00%0.31 ~ 0.32
3201155.00%0.32 ~ 0.33
4201785.00%0.33 ~ 0.34
5201785.00%0.34 ~ 0.35
6201050.00%0.35 ~ 0.35
72021105.00%0.35 ~ 0.36
8201260.00%0.36 ~ 0.37
9201680.00%0.37 ~ 0.38
10201050.00%0.39 ~ 0.46

LR + WOE 评分卡 — OOT 集 坏样本率分档

分组样本数坏样本坏样本率分数范围
1201365.00%0.28 ~ 0.31
2201785.00%0.31 ~ 0.32
3201260.00%0.32 ~ 0.33
420840.00%0.33 ~ 0.33
5201050.00%0.33 ~ 0.34
62020100.00%0.34 ~ 0.35
7201365.00%0.35 ~ 0.36
820840.00%0.36 ~ 0.37
9201365.00%0.37 ~ 0.39
10201785.00%0.39 ~ 0.49

5. 评分卡

变量分箱WOE系数分数
age(-inf, 33.0]-0.35550.581431
age(33.0, 47.0]-0.20240.581429
age(47.0, 52.0]-0.24570.581429
age(52.0, inf]-0.27150.581430
genderF-0.51240.034726
genderM-0.66860.034726
education大专-0.55090.168228
education本科-0.42280.168227
education硕士及以上-0.15800.168226
education高中及以下-0.40970.168227
marital_status丧偶-0.00120.067125
marital_status已婚-0.51130.067126
marital_status未婚-0.65040.067126
marital_status离异-0.17520.067126
employment_years(-inf, 2.0]-0.07990.156726
employment_years(2.0, 5.0]-0.30630.156727
employment_years(5.0, inf]-0.41970.156727
credit_score(-inf, 692.3]-0.3918-0.245022
credit_score(692.3, inf]-0.3758-0.245023
existing_loans(-inf, 1.0]-0.43260.505532
existing_loans(1.0, 1.857142857142776]0.00000.505525
existing_loans(1.857142857142776, inf]-0.32530.505530
debt_ratio(-inf, 0.0609]-0.05830.992827
debt_ratio(0.0609, 0.091]-0.24900.992832
debt_ratio(0.091, 0.122]-0.06670.992827
debt_ratio(0.122, 0.152]-0.20860.992831
debt_ratio(0.152, 0.2922]-0.26880.992833
debt_ratio(0.2922, inf]-0.33580.992835
months_since_last_loan(-inf, 12.0]-0.21140.322127
months_since_last_loan(12.0, inf]-0.47010.322130
loan_term(-inf, 24.0]-0.4383-1.102511
loan_term(24.0, inf]-0.2389-1.102518
purpose其他-0.11380.370026
purpose医疗-0.25580.370028
purpose周转-0.53050.370031
purpose家装-0.19470.370027
purpose教育-0.21600.370028
purpose日常消费-0.41950.370030
channelAPP-0.5501-0.254121
channelWEB-0.4063-0.254122
channel合作渠道-0.3096-0.254123
channel线下-0.2665-0.254123
region东北-0.25750.660730
region华东-0.24980.660730
region华中-0.26720.660730
region华北-0.22880.660730
region华南-0.31310.660731
region西南-0.32100.660731
region西部-0.30300.660731
city_tier(-inf, 3.0]-0.3838-0.102324
city_tier(3.0, inf]-0.3825-0.102324
app_sessions_30d(-inf, 9.0]-0.4024-0.243922
app_sessions_30d(9.0, inf]-0.3497-0.243923
clicks_before_apply(-inf, 21.0]-0.4595-0.454319
clicks_before_apply(21.0, inf]-0.1484-0.454323
time_on_page_sec(-inf, 483.0]-0.47460.647434
time_on_page_sec(483.0, inf]-0.19390.647429
fpd30(-inf, inf]-0.4840-0.130523
constant_flag(-inf, inf]-0.4840-0.130523