Analytics: April 2016

The analysis was generated using SAS Studio and the data was sourced from the UCI Machine Learning Repository. The objective was to use random forests as a data mining technique to predict the probability of a person defaulting on their credit card payments. This evaluation is useful in credit scoring models.

ATTRIBUTE INFORMATION:

There are 30,000 observations.
The binary variable, default_payment_next_month (Yes=2, No=1) is the response variable.
There are 23 explanatory variables:
LIMIT_BAL: Amount of the given credit (NT dollar): it includes both the individual consumer credit and his/her family (supplementary) credit.
SEX: (1 = male; 2 = female).
EDUCATION : (1 = graduate school; 2 = university; 3 = high school; 4 = others).
MARRIAGE: (1 = married; 2 = single; 3 = others).
AGE: (year).
History of past payment: REPAY_SEP, REPAY_AUG, REPAY_JUL, REPAY_JUN, REPAY_MAY, REPAY_APR. The measurement scale for the repayment status is: -1 = pay duly; 1 = payment delay for one month; 2 = payment delay for two months; . . .; 8 = payment delay for eight months; 9 = payment delay for nine months and above.
Amount of bill statement: BILL_SEP, BILL_AUG, BILL_JUL, BILL_JUN, BILL_APR, BILL_MAY
Amount of previous payment: PAY_AMT_SEP, PAY_AMT_AUG, PAY_AMT_JUL, PAY_AMT_JUN, PAY_AMT_MAY, PAY_AMT_APR;

SAS CODE

/* Source File: default of credit card clients.xls */

/* Source Path: /home/mst07221 */

%web_drop_table(WORK.IMPORT2);

FILENAME REFFILE "/home/mst07221/default of credit card clients.xls" TERMSTR=CR;

PROC IMPORT DATAFILE=REFFILE

DBMS=XLS

OUT=WORK.IMPORT2;

GETNAMES=YES;

RUN;

PROC SORT DATA=WORK.IMPORT2;

BY LIMIT_BAL;

ods graphics on;

PROC HPFOREST;

target default_payment_next_month/level=nominal;

input

SEX EDUCATION MARRIAGE REPAY_SEP REPAY_AUG REPAY_JUL REPAY_JUN REPAY_MAY REPAY_APR /level=nominal;

input

LIMIT_BAL AGE BILL_SEP BILL_AUG BILL_JUL BILL_JUN BILL_APR BILL_MAY PAY_AMT_SEP PAY_AMT_AUG PAY_AMT_JUL PAY_AMT_JUN PAY_AMT_MAY PAY_AMT_APR /level=interval;

RUN;

%web_open_table(WORK.IMPORT2);

RESULTS

The HPFOREST Procedure

Performance Information
Execution Mode	Single-Machine
Number of Threads	2

Data Access Information
Data	Engine	Role	Path
WORK.IMPORT2	V9	Input	On Client

Model Information
Parameter	Value
Variables to Try	5	(Default)
Maximum Trees	100	(Default)
Inbag Fraction	0.6	(Default)
Prune Fraction	0	(Default)
Prune Threshold	0.1	(Default)
Leaf Fraction	0.00001	(Default)
Leaf Size Setting	1	(Default)
Leaf Size Used	1
Category Bins	30	(Default)
Interval Bins	100
Minimum Category Size	5	(Default)
Node Size	100000	(Default)
Maximum Depth	20	(Default)
Alpha	1	(Default)
Exhaustive	5000	(Default)
Rows of Sequence to Skip	5	(Default)
Split Criterion	.	Gini
Preselection Method	.	Loh
Missing Value Handling	.	Valid value

Number of Observations
Type	N
Number of Observations Read	30000
Number of Observations Used	30000

Baseline Fit Statistics
Statistic	Value
Average Square Error	0.172
Misclassification Rate	0.221
Log Loss	0.528

Fit Statistics
Number of Trees	Number of Leaves	Average Square Error (Train)	Average Square Error (OOB)	Misclassification Rate (Train)	Misclassification Rate (OOB)	Log Loss (Train)	Log Loss (OOB)
1	3294	0.1139	0.237	0.1269	0.249	1.966	4.767
2	6520	0.0798	0.225	0.1135	0.244	0.513	4.096
3	9603	0.0687	0.212	0.0879	0.239	0.280	3.443
4	12768	0.0632	0.202	0.0822	0.234	0.227	2.954
5	15775	0.0603	0.193	0.0758	0.229	0.213	2.520
6	18789	0.0584	0.186	0.0736	0.223	0.208	2.191
7	21727	0.0570	0.179	0.0713	0.219	0.206	1.893
8	24756	0.0559	0.174	0.0692	0.215	0.205	1.644
9	27565	0.0553	0.169	0.0672	0.210	0.203	1.445
10	30525	0.0548	0.165	0.0660	0.208	0.203	1.275
11	33342	0.0545	0.162	0.0652	0.206	0.203	1.142
12	36112	0.0544	0.160	0.0654	0.205	0.203	1.037
13	39169	0.0540	0.157	0.0657	0.202	0.203	0.942
14	42015	0.0538	0.155	0.0648	0.201	0.203	0.862
15	44959	0.0536	0.154	0.0642	0.200	0.203	0.811
16	47860	0.0534	0.152	0.0641	0.199	0.203	0.754
17	50907	0.0531	0.151	0.0636	0.196	0.202	0.714
18	53923	0.0528	0.150	0.0636	0.196	0.202	0.674
19	56981	0.0526	0.149	0.0628	0.194	0.202	0.645
20	60094	0.0523	0.148	0.0630	0.195	0.201	0.618
21	63098	0.0521	0.148	0.0624	0.194	0.201	0.596
22	66296	0.0517	0.147	0.0618	0.194	0.200	0.580
23	69211	0.0516	0.146	0.0621	0.193	0.200	0.559
24	72437	0.0512	0.146	0.0614	0.192	0.199	0.550
25	75734	0.0509	0.145	0.0606	0.192	0.198	0.540
26	78841	0.0507	0.145	0.0604	0.192	0.198	0.530
27	81858	0.0505	0.145	0.0601	0.191	0.197	0.521
28	84911	0.0505	0.144	0.0599	0.192	0.197	0.515
29	87702	0.0505	0.144	0.0604	0.191	0.198	0.509
30	90681	0.0505	0.144	0.0603	0.191	0.198	0.505
31	93747	0.0504	0.143	0.0600	0.191	0.197	0.497
32	96588	0.0504	0.143	0.0602	0.189	0.198	0.492
33	99778	0.0502	0.143	0.0592	0.190	0.197	0.489
34	102728	0.0502	0.143	0.0595	0.189	0.197	0.486
35	105948	0.0500	0.142	0.0595	0.188	0.197	0.482
36	108908	0.0500	0.142	0.0590	0.189	0.197	0.478
37	111980	0.0499	0.142	0.0589	0.189	0.196	0.475
38	114907	0.0499	0.142	0.0592	0.188	0.196	0.473
39	117971	0.0498	0.142	0.0588	0.188	0.196	0.471
40	120864	0.0498	0.141	0.0590	0.188	0.197	0.470
41	123726	0.0499	0.141	0.0584	0.187	0.197	0.468
42	126662	0.0498	0.141	0.0586	0.187	0.197	0.466
43	129557	0.0498	0.141	0.0591	0.187	0.197	0.466
44	132304	0.0499	0.141	0.0593	0.188	0.197	0.461
45	135209	0.0499	0.141	0.0589	0.187	0.197	0.460
46	138021	0.0499	0.140	0.0592	0.188	0.197	0.458
47	141315	0.0498	0.140	0.0591	0.187	0.197	0.457
48	144402	0.0497	0.140	0.0588	0.187	0.197	0.455
49	147425	0.0497	0.140	0.0587	0.187	0.196	0.453
50	150491	0.0496	0.140	0.0585	0.187	0.196	0.452
51	153679	0.0495	0.140	0.0585	0.187	0.196	0.451
52	156685	0.0495	0.140	0.0586	0.186	0.196	0.450
53	159472	0.0496	0.140	0.0588	0.187	0.197	0.450
54	162672	0.0495	0.140	0.0584	0.187	0.196	0.450
55	165562	0.0495	0.139	0.0579	0.186	0.196	0.449
56	168637	0.0495	0.139	0.0580	0.187	0.196	0.449
57	171394	0.0495	0.139	0.0583	0.186	0.196	0.448
58	174267	0.0495	0.139	0.0584	0.186	0.196	0.448
59	177350	0.0495	0.139	0.0583	0.186	0.196	0.447
60	180405	0.0494	0.139	0.0583	0.186	0.196	0.447
61	183345	0.0494	0.139	0.0580	0.186	0.196	0.447
62	186120	0.0494	0.139	0.0585	0.186	0.196	0.446
63	188734	0.0495	0.139	0.0585	0.186	0.196	0.446
64	191563	0.0495	0.139	0.0587	0.186	0.196	0.446
65	194619	0.0495	0.139	0.0586	0.186	0.196	0.446
66	197498	0.0495	0.139	0.0588	0.185	0.196	0.445
67	200432	0.0495	0.138	0.0587	0.185	0.196	0.444
68	203318	0.0495	0.138	0.0592	0.186	0.197	0.444
69	206392	0.0495	0.138	0.0588	0.186	0.196	0.444
70	209681	0.0494	0.138	0.0585	0.186	0.196	0.444
71	212692	0.0494	0.138	0.0583	0.186	0.196	0.443
72	215727	0.0494	0.138	0.0580	0.185	0.196	0.442
73	218663	0.0494	0.138	0.0579	0.185	0.196	0.442
74	221696	0.0493	0.138	0.0580	0.185	0.196	0.442
75	224667	0.0493	0.138	0.0581	0.185	0.196	0.442
76	227649	0.0493	0.138	0.0582	0.184	0.196	0.442
77	230608	0.0493	0.138	0.0583	0.184	0.196	0.441
78	233755	0.0493	0.138	0.0579	0.184	0.196	0.441
79	236390	0.0493	0.138	0.0580	0.184	0.196	0.441
80	239439	0.0493	0.138	0.0577	0.184	0.196	0.441
81	242146	0.0493	0.138	0.0580	0.184	0.196	0.441
82	245104	0.0493	0.138	0.0577	0.184	0.196	0.440
83	248185	0.0493	0.138	0.0580	0.184	0.196	0.440
84	251073	0.0493	0.138	0.0580	0.184	0.196	0.440
85	253984	0.0493	0.138	0.0577	0.184	0.196	0.440
86	256912	0.0493	0.138	0.0578	0.184	0.196	0.440
87	260062	0.0492	0.138	0.0576	0.183	0.196	0.440
88	263180	0.0492	0.138	0.0577	0.184	0.196	0.440
89	266373	0.0491	0.138	0.0576	0.184	0.196	0.440
90	269247	0.0491	0.138	0.0576	0.184	0.196	0.440
91	272239	0.0491	0.138	0.0576	0.184	0.196	0.440
92	275375	0.0490	0.138	0.0574	0.184	0.195	0.439
93	278332	0.0491	0.138	0.0571	0.185	0.196	0.439
94	281349	0.0491	0.137	0.0573	0.184	0.196	0.439
95	284273	0.0490	0.137	0.0574	0.185	0.196	0.439
96	287059	0.0491	0.137	0.0578	0.184	0.196	0.439
97	289819	0.0491	0.137	0.0577	0.184	0.196	0.439
98	292997	0.0491	0.137	0.0577	0.184	0.196	0.439
99	295907	0.0491	0.137	0.0576	0.184	0.196	0.439
100	298931	0.0491	0.137	0.0577	0.184	0.196	0.439

Loss Reduction Variable Importance
Variable	Number of Rules	Gini	OOB Gini	Margin	OOB Margin
REPAY_SEP	4520	0.033755	0.02978	0.067509	0.06421
REPAY_AUG	3243	0.012217	0.01002	0.024434	0.02243
REPAY_JUL	3164	0.008982	0.00674	0.017965	0.01581
REPAY_JUN	3369	0.005476	0.00311	0.010953	0.00893
REPAY_MAY	3577	0.004900	0.00244	0.009800	0.00777
REPAY_APR	3738	0.004577	0.00182	0.009154	0.00685
SEX	6255	0.002262	-0.00184	0.004524	0.00045
MARRIAGE	6520	0.002317	-0.00221	0.004635	0.00038
EDUCATION	6383	0.002914	-0.00266	0.005829	0.00053
PAY_AMT_SEP	16773	0.013924	-0.00935	0.027848	0.00424
BILL_APR	14566	0.010728	-0.00981	0.021456	0.00069
BILL_JUN	15011	0.011076	-0.01002	0.022153	0.00076
LIMIT_BAL	19721	0.014944	-0.01024	0.029889	0.00483
PAY_AMT_AUG	18266	0.014662	-0.01062	0.029323	0.00383
BILL_JUL	15948	0.011740	-0.01064	0.023479	0.00094
BILL_SEP	16241	0.012512	-0.01081	0.025024	0.00137
BILL_AUG	16691	0.012304	-0.01088	0.024608	0.00121
BILL_MAY	15693	0.011328	-0.01111	0.022657	0.00011
PAY_AMT_JUL	19324	0.013843	-0.01221	0.027686	0.00152
PAY_AMT_JUN	21109	0.014101	-0.01352	0.028202	0.00043
AGE	19789	0.014175	-0.01394	0.028349	0.00049
PAY_AMT_MAY	22755	0.014860	-0.01474	0.029721	-0.00008
PAY_AMT_APR	26175	0.016937	-0.01669	0.033874	0.00018

SUMMARY OF FINDINGS

All the 30,000 observations were used in the model because there were no missing values in the predictor variables.

The ‘Variables to Try’ parameter indicates that 5 of the 23 explanatory variables were randomly selected to be considered for a splitting rule.

PROC HPFOREST first computes baseline statistics without using a model. The Baseline Fit Statistics table shows a baseline misclassification rate of 0.221 because that is the proportion of observations for which default_payment_next_month =2 (Yes).

The Fit Statistics table shows that as the number of trees increases, the fit statistics improve (decrease) at first and then level off and fluctuate in a small range. They decrease from 0.1139 to 0.0491.

The table also provides an alternative estimate of average square error (ASE) and misclassification rate - the out-of-bag (OOB) estimate. This is a convenient substitute for an estimate that is based on test data and is a less biased estimate of how the model will perform on future data. The ASE OOB is worse (larger) than the estimate that evaluates all observations on all trees. The OOB misclassification rate decreases to values that are below the baseline misclassification rate (good model).

The Loss Reduction Variable Importance table shows that each measure is computed twice: once on training data and once on out-of-bag data. As with fit statistics, the out-of-bag estimates are less biased. The rows are sorted by the OOB Gini measure, which is a more stringent measure than the OOB margin measure. The OOB Gini column is negative for 17 of the 23 variables, and the OOB margin column is negative for one of the 23 variables.

We can conclude that the REPAY_SEP value is the most important predictor of whether one will default on their next month's credit card payment.

Reference

Yeh, I. C., & Lien, C. H. (2009). The comparisons of data mining techniques for the predictive accuracy of probability of default of credit card clients. Expert Systems with Applications, 36(2), 2473-2480.

Analytics

Saturday, April 30, 2016

Machine Learning Data Analysis: Random Forests