阿里天池竞赛

【教学赛】金融数据分析赛题1：银行客户认购产品预测

课程论文

数据

To DO: 预测用户是否进行购买产品

数据解释

字段	说明
age	年龄
job	职业：admin, unknown, unemployed, management…
marital	婚姻：married, divorced, single
education	受教育程度
default	信用卡是否有违约: yes or no
housing	是否有房贷: yes or no
contact	联系方式：unknown, telephone, cellular
month	上一次联系的月份：jan, feb, mar, …
day_of_week	上一次联系的星期几：mon, tue, wed, thu, fri
duration	上一次联系的时长（秒）
campaign	活动期间联系客户的次数
pdays	上一次与客户联系后的间隔天数
previous	在本次营销活动前，与客户联系的次数
poutcome	之前营销活动的结果：unknown, other, failure, success
emp_var_rate	就业变动率（季度指标）
cons_price_index	消费者价格指数（月度指标）
cons_conf_index	消费者信心指数（月度指标）
lending_rate3m	银行同业拆借率 3个月利率（每日指标）
nr_employed	雇员人数（季度指标）
subscribe	客户是否进行购买：yes 或 no

训练数据

共有22500条记录，每条记录有21个字段（包括subscribe）

id	age	job	marital	education	default	housing	loan	contact	month	day_of_week	duration	campaign	pdays	previous	poutcome	emp_var_rate	cons_price_index	cons_conf_index	lending_rate3m	nr_employed	subscribe
1	51	admin.	divorced	professional.course	no	yes	yes	cellular	aug	mon	4621	1	112	2	failure	1.4	90.81	-35.53	0.69	5219.74	no
2	50	services	married	high.school	unknown	yes	no	cellular	may	mon	4715	1	412	2	nonexistent	-1.8	96.33	-40.58	4.05	4974.79	yes
3	48	blue-collar	divorced	basic.9y	no	no	no	cellular	apr	wed	171	0	1027	1	failure	-1.8	96.33	-44.74	1.5	5022.61	no
4	26	entrepreneur	single	high.school	yes	yes	yes	cellular	aug	fri	359	26	998	0	nonexistent	1.4	97.08	-35.55	5.11	5222.87	yes
5	45	admin.	single	university.degree	no	no	no	cellular	nov	tue	3178	1	240	4	success	-3.4	89.82	-33.83	1.17	4884.7	no

测试数据

共有7500条记录，每条记录有20个字段（去除了subscribe）

id	age	job	marital	education	default	housing	loan	contact	month	day_of_week	duration	campaign	pdays	previous	poutcome	emp_var_rate	cons_price_index	cons_conf_index	lending_rate3m	nr_employed
22501	35	technician	single	professional.course	no	yes	yes	cellular	aug	mon	3295	1	476	0	success	1.4	95.37	-33.04	3.63	5204.54
22502	26	admin.	single	high.school	no	yes	no	cellular	may	thu	2872	1	166	2	success	-1.8	91.75	-44.42	3.16	4924.78
22503	44	blue-collar	married	basic.6y	no	no	no	telephone	may	mon	268	3	968	0	nonexistent	1.1	89.67	-36.9	5.04	4947.02
22504	36	blue-collar	married	basic.9y	no	yes	no	telephone	nov	thu	30	1	432	5	success	-0.1	89.87	-41.66	3.27	5203.33
22505	41	blue-collar	married	basic.4y	no	yes	no	telephone	may	thu	1670	1	944	3	success	1.1	97.64	-36.32	3.95	4992.02

提交格式

id	subscribe
22501	no
22502	no
22503	no
22504	no
22505	no

数据编码

数据编码分为两种，1是将所有数据全部编码为0-1范围内，2是采用独热编码

• 编码方式1：全部编码到0-1内

字段	字段值	编码
job	admin. services blue-collar entrepreneur management technician housemaid self-employed unemployed retired student unknown	0.00 0.09 0.18 0.27 0.36 0.45 0.55 0.64 0.73 0.82 0.91 1.00
marital	divorced married single unknown	0.00 0.33 0.67 1.00
education	professional.course high.school basic.9y university.degree unknown basic.4y basic.6y illiterate	0.00 0.14 0.29 0.43 0.57 0.71 0.86 1.00
default	no unknown yes	0.00 0.50 1.00
housing	yes no unknown	0.00 0.50 1.00
loan	yes no unknown	0.00 0.50 1.00
contact	cellular telephone	0.00 1.00
month	aug may apr nov jul jun oct dec sep mar	0.00 0.11 0.22 0.33 0.44 0.56 0.67 0.78 0.89 1.00
day_of_week	mon wed fri tue thu	0.00 0.25 0.50 0.75 1.00
poutcome	failure nonexistent success	0.00 0.50 1.00
subscribe	no yes	0.00 1.00

• 编码方式2：独热编码

字段	字段值	编码
job	admin. services blue-collar entrepreneur management technician housemaid self-employed unemployed retired student unknown	[1. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.] [0. 1. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.] [0. 0. 1. 0. 0. 0. 0. 0. 0. 0. 0. 0.] [0. 0. 0. 1. 0. 0. 0. 0. 0. 0. 0. 0.] [0. 0. 0. 0. 1. 0. 0. 0. 0. 0. 0. 0.] [0. 0. 0. 0. 0. 1. 0. 0. 0. 0. 0. 0.] [0. 0. 0. 0. 0. 0. 1. 0. 0. 0. 0. 0.] [0. 0. 0. 0. 0. 0. 0. 1. 0. 0. 0. 0.] [0. 0. 0. 0. 0. 0. 0. 0. 1. 0. 0. 0.] [0. 0. 0. 0. 0. 0. 0. 0. 0. 1. 0. 0.] [0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 1. 0.] [0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 1.]
marital	divorced married single unknown	[1. 0. 0. 0.] [0. 1. 0. 0.] [0. 0. 1. 0.] [0. 0. 0. 1.]
education	professional.course high.school basic.9y university.degree unknown basic.4y basic.6y illiterate	[1. 0. 0. 0. 0. 0. 0. 0.] [0. 1. 0. 0. 0. 0. 0. 0.] [0. 0. 1. 0. 0. 0. 0. 0.] [0. 0. 0. 1. 0. 0. 0. 0.] [0. 0. 0. 0. 1. 0. 0. 0.] [0. 0. 0. 0. 0. 1. 0. 0.] [0. 0. 0. 0. 0. 0. 1. 0.] [0. 0. 0. 0. 0. 0. 0. 1.]
default	no unknown yes	[1. 0. 0.] [0. 1. 0.] [0. 0. 1.]
housing	yes no unknown	[1. 0. 0.] [0. 1. 0.] [0. 0. 1.]
loan	yes no unknown	[1. 0. 0.] [0. 1. 0.] [0. 0. 1.]
contact	cellular telephone	[1. 0.] [0. 1.]
month	aug may apr nov jul jun oct dec sep mar	[1. 0. 0. 0. 0. 0. 0. 0. 0. 0.] [0. 1. 0. 0. 0. 0. 0. 0. 0. 0.] [0. 0. 1. 0. 0. 0. 0. 0. 0. 0.] [0. 0. 0. 1. 0. 0. 0. 0. 0. 0.] [0. 0. 0. 0. 1. 0. 0. 0. 0. 0.] [0. 0. 0. 0. 0. 1. 0. 0. 0. 0.] [0. 0. 0. 0. 0. 0. 1. 0. 0. 0.] [0. 0. 0. 0. 0. 0. 0. 1. 0. 0.] [0. 0. 0. 0. 0. 0. 0. 0. 1. 0.] [0. 0. 0. 0. 0. 0. 0. 0. 0. 1.]
day_of_week	mon wed fri tue thu	[1. 0. 0. 0. 0.] [0. 1. 0. 0. 0.] [0. 0. 1. 0. 0.] [0. 0. 0. 1. 0.] [0. 0. 0. 0. 1.]
poutcome	failure nonexistent success	[1. 0. 0.] [0. 1. 0.] [0. 0. 1.]
subscribe	no yes	[1. 0.] [0. 1.]

模型结构

全连接神经网络，5层，每层2048节点。可选：ReLU激活函数、batch normlization、onehot编码。

结果

MSE & Accuracy during training process. Training settings: batch_size = 2500, epochs = 10.

• Encode strings into a value
  • 1-1 normalization & ReLU，提交得分：0.9117
  • 1-2 normalization & ReLU，提交得分：0.9007
  • 1-3 normalization & Sigmoid，提交得分：0.7845
  • 1-4 normalization & Sigmoid，提交得分：0.9224
• Onehot
  • 2-1 normalization & ReLU，提交得分：0.9029
  • 2-2 normalization & ReLU，提交得分：0
  • 2-3 normalization & Sigmoid，提交得分：0.9217
  • 2-4 normalization & Sigmoid，提交得分：0

MSE (1-1 normalization & ReLU) 0.01	Accuracy (1-1 normalization & ReLU) 98.88%
MSE (1-2 normalization & ReLU) 0.01	Accuracy (1-2 normalization & ReLU) 98.60%
MSE (1-3 normalization & Sigmoid) 0.07	Accuracy (1-3 normalization & Sigmoid) 91.40%
MSE (1-4 normalization & Sigmoid) 0.13	Accuracy (1-4 normalization & Sigmoid) 86.84%
MSE 2-1 normalization & ReLU 0.01	Accuracy 2-1 normalization & ReLU 98.84%
MSE 2-2 normalization & ReLU 0.13	Accuracy 2-2 normalization & ReLU 87.24%
MSE 2-3 normalization & Sigmoid 0.10	Accuracy 2-3 normalization & Sigmoid 94.48%
MSE 2-4 normalization & Sigmoid 0.13	Accuracy 2-4 normalization & Sigmoid 86.68%

参赛队伍

我组件了一支单人队伍，队伍名称是finishtheless