General Info

Home
News
CEO
Mail us
Contact
Links

Product Info

Shih Data Miner
Apps & Services
Screenshots
Technical Info
Download
ChangeLog
Manual
License
Recent users

Income rules
Motors case

Publications

Classification methods

Predicting income

This illustrates the type of decision rules you get with a tree model induced them from the training data. Shih data miner allows to manually modify the rules and see the impact in the model. This is usefull to adjust them to the company's policies.

The goal is to make a model to classify between 2 types of people, the ones that win more than 50 thousand dollars, and the ones who don't. This study was needed for the US tax office in order to identify the people who had high probability of winning above that amount in order to control beforehand if they where paying their taxes or not.

The following rules are the most important ones in order to identify between the two types of income. Note that the percentage of people wining more than 50K was 23.5% in the training data, therefore a node with more percentage than 23.5% will be considered as a node >50K. The rules are ordered in decreasing sequence from the higher to the lower income probability:

[RULE: 15, Income --> >50K.]

IF capital-gain > 5095.5
AND education-num <= 9.5
AND relationship IN (' Wife',' Husband')
THEN Income = >50K. with probability : 97.64706%

[RULE: 24, Income --> >50K.]

IF capital-gain > 7055.5
AND relationship IN (' Other-relative',' Unmarried',' Not-in-family',' Own-child')
THEN Income = >50K. with probability : 96.33028%

[RULE: 14, Income --> >50K.]

IF capital-loss > 1794.0
AND occupation IN (' Armed-Forces',' Handlers-cleaners',' Transport-moving',' Priv-house-serv',' Sales',' Adm-clerical',' Craft-repair',' Prof-specialty',' ?',' Protective-serv',' Machine-op-inspct')
AND age > 26.5
AND hours-per-week > 36.5
AND 8.5 < education-num <= 9.5
AND capital-gain <= 5095.5
AND relationship IN (' Wife',' Husband')
THEN Income = >50K. with probability : 72.22223%

[RULE: 18, Income --> >50K.]

IF age > 27.5
AND education-num > 9.5
AND relationship IN (' Wife',' Husband')
THEN Income = >50K. with probability : 61.43148%

[RULE: 7, Income --> >50K.]

IF occupation IN (' Tech-support',' Exec-managerial')
AND age > 26.5
AND hours-per-week > 36.5
AND 8.5 < education-num <= 9.5
AND capital-gain <= 5095.5
AND relationship IN (' Wife',' Husband')
THEN Income = >50K. with probability : 52.499996%

[RULE: 22, Income --> <=50K.]

IF hours-per-week > 40.5
AND age > 33.5
AND education-num > 12.5
AND capital-gain <= 4668.5
AND relationship IN (' Other-relative',' Unmarried',' Not-in-family',' Own-child')
THEN Income = <=50K. with probability : 62.15139%

[RULE: 13, Income --> <=50K.]

IF workclass IN (' Never-worked',' Without-pay',' Self-emp-inc',' State-gov',' Federal-gov',' Self-emp-not-inc',' Local-gov',' Private')
AND native-country IN (' France',' Outlying-US(Guam-USVI-etc)',' Trinadad&Tobago',' Greece',' Hong',' Hungary',' Yugoslavia',' Ecuador',' Jamaica',' Scotland',' Iran',' Honduras',' Nicaragua',' China',' Canada',' Italy',' Taiwan',' Cuba',' England',' Laos',' Poland',' Cambodia',' India',' Japan',' Columbia',' South',' Vietnam',' Puerto-Rico',' El-Salvador',' Haiti',' Thailand',' Philippines',' Germany',' Ireland',' Dominican-Republic',' Guatemala',' Peru',' ?',' United-States')
AND age > 35.5
AND capital-loss <= 1794.0
AND occupation IN (' Armed-Forces',' Handlers-cleaners',' Transport-moving',' Priv-house-serv',' Sales',' Adm-clerical',' Craft-repair',' Prof-specialty',' ?',' Protective-serv',' Machine-op-inspct')
AND hours-per-week > 36.5
AND 8.5 < education-num <= 9.5
AND capital-gain <= 5095.5
AND relationship IN (' Wife',' Husband')
THEN Income = <=50K. with probability : 64.29549%

[RULE: 9, Income --> <=50K.]

IF occupation IN (' Armed-Forces',' Handlers-cleaners',' Priv-house-serv',' Sales',' Craft-repair',' Prof-specialty',' ?',' Protective-serv',' Machine-op-inspct')
AND fnlwgt <= 185452.0
AND 26.5 < age <= 35.5
AND capital-loss <= 1794.0
AND hours-per-week > 36.5
AND 8.5 < education-num <= 9.5
AND capital-gain <= 5095.5
AND relationship IN (' Wife',' Husband')
THEN Income = <=50K. with probability : 66.666664%

[RULE: 3, Income --> <=50K.]

IF age > 41.5
AND hours-per-week > 40.5
AND education-num <= 8.5
AND capital-gain <= 5095.5
AND relationship IN (' Wife',' Husband')
THEN Income = <=50K. with probability : 69.60784%

[RULE: 23, Income --> <=50K.]

IF 4668.5 < capital-gain <= 7055.5
AND relationship IN (' Other-relative',' Unmarried',' Not-in-family',' Own-child')
THEN Income = <=50K. with probability : 72.34042%

[RULE: 17, Income --> <=50K.]

IF education IN (' Preschool',' 1st-4th',' 12th',' 9th',' Assoc-voc',' 5th-6th',' Masters',' Bachelors',' 7th-8th',' Prof-school',' 10th',' Assoc-acdm',' HS-grad',' 11th')
AND age <= 27.5
AND education-num > 9.5
AND relationship IN (' Wife',' Husband')
THEN Income = <=50K. with probability : 73.74999%

[RULE: 10, Income --> <=50K.]

IF fnlwgt > 185452.0
AND 26.5 < age <= 35.5
AND capital-loss <= 1794.0
AND occupation IN (' Armed-Forces',' Handlers-cleaners',' Transport-moving',' Priv-house-serv',' Sales',' Adm-clerical',' Craft-repair',' Prof-specialty',' ?',' Protective-serv',' Machine-op-inspct')
AND hours-per-week > 36.5
AND 8.5 < education-num <= 9.5
AND capital-gain <= 5095.5
AND relationship IN (' Wife',' Husband')
THEN Income = <=50K. with probability : 84.210526%

[RULE: 2, Income --> <=50K.]

IF age <= 41.5
AND hours-per-week > 40.5
AND education-num <= 8.5
AND capital-gain <= 5095.5
AND relationship IN (' Wife',' Husband')
THEN Income = <=50K. with probability : 84.507034%

[RULE: 8, Income --> <=50K.]

IF occupation IN (' Transport-moving',' Adm-clerical')
AND fnlwgt <= 185452.0
AND 26.5 < age <= 35.5
AND capital-loss <= 1794.0
AND hours-per-week > 36.5
AND 8.5 < education-num <= 9.5
AND capital-gain <= 5095.5
AND relationship IN (' Wife',' Husband')
THEN Income = <=50K. with probability : 85.0%

[RULE: 12, Income --> <=50K.]

IF workclass IN (' ?')
AND native-country IN (' France',' Outlying-US(Guam-USVI-etc)',' Trinadad&Tobago',' Greece',' Hong',' Hungary',' Yugoslavia',' Ecuador',' Jamaica',' Scotland',' Iran',' Honduras',' Nicaragua',' China',' Canada',' Italy',' Taiwan',' Cuba',' England',' Laos',' Poland',' Cambodia',' India',' Japan',' Columbia',' South',' Vietnam',' Puerto-Rico',' El-Salvador',' Haiti',' Thailand',' Philippines',' Germany',' Ireland',' Dominican-Republic',' Guatemala',' Peru',' ?',' United-States')
AND age > 35.5
AND capital-loss <= 1794.0
AND occupation IN (' Armed-Forces',' Handlers-cleaners',' Transport-moving',' Priv-house-serv',' Sales',' Adm-clerical',' Craft-repair',' Prof-specialty',' ?',' Protective-serv',' Machine-op-inspct')
AND hours-per-week > 36.5
AND 8.5 < education-num <= 9.5
AND capital-gain <= 5095.5
AND relationship IN (' Wife',' Husband')
THEN Income = <=50K. with probability : 86.206894%

[RULE: 21, Income --> <=50K.]

IF hours-per-week <= 40.5
AND age > 33.5
AND education-num > 12.5
AND capital-gain <= 4668.5
AND relationship IN (' Other-relative',' Unmarried',' Not-in-family',' Own-child')
THEN Income = <=50K. with probability : 86.27002%

[RULE: 5, Income --> <=50K.]

IF hours-per-week <= 36.5
AND occupation IN (' Armed-Forces',' Handlers-cleaners',' Transport-moving',' Priv-house-serv',' Sales',' Tech-support',' Exec-managerial',' Adm-clerical',' Craft-repair',' Prof-specialty',' ?',' Protective-serv',' Machine-op-inspct')
AND 8.5 < education-num <= 9.5
AND capital-gain <= 5095.5
AND relationship IN (' Wife',' Husband')
THEN Income = <=50K. with probability : 86.36364%

[RULE: 4, Income --> <=50K.]

IF occupation IN (' Other-service',' Farming-fishing')
AND 8.5 < education-num <= 9.5
AND capital-gain <= 5095.5
AND relationship IN (' Wife',' Husband')
THEN Income = <=50K. with probability : 89.44444%

[RULE: 6, Income --> <=50K.]

IF age <= 26.5
AND hours-per-week > 36.5
AND occupation IN (' Armed-Forces',' Handlers-cleaners',' Transport-moving',' Priv-house-serv',' Sales',' Tech-support',' Exec-managerial',' Adm-clerical',' Craft-repair',' Prof-specialty',' ?',' Protective-serv',' Machine-op-inspct')
AND 8.5 < education-num <= 9.5
AND capital-gain <= 5095.5
AND relationship IN (' Wife',' Husband')
THEN Income = <=50K. with probability : 90.10989%

[RULE: 16, Income --> <=50K.]

IF education IN (' Doctorate',' Some-college')
AND age <= 27.5
AND education-num > 9.5
AND relationship IN (' Wife',' Husband')
THEN Income = <=50K. with probability : 92.473114%

[RULE: 1, Income --> <=50K.]

IF hours-per-week <= 40.5
AND education-num <= 8.5
AND capital-gain <= 5095.5
AND relationship IN (' Wife',' Husband')
THEN Income = <=50K. with probability : 93.647545%

[RULE: 20, Income --> <=50K.]

IF age <= 33.5
AND education-num > 12.5
AND capital-gain <= 4668.5
AND relationship IN (' Other-relative',' Unmarried',' Not-in-family',' Own-child')
THEN Income = <=50K. with probability : 94.09091%

[RULE: 19, Income --> <=50K.]

IF education-num <= 12.5
AND capital-gain <= 4668.5
AND relationship IN (' Other-relative',' Unmarried',' Not-in-family',' Own-child')
THEN Income = <=50K. with probability : 97.81643%

[RULE: 11, Income --> <=50K.]

IF native-country IN (' Portugal',' Mexico')
AND age > 35.5
AND capital-loss <= 1794.0
AND occupation IN (' Armed-Forces',' Handlers-cleaners',' Transport-moving',' Priv-house-serv',' Sales',' Adm-clerical',' Craft-repair',' Prof-specialty',' ?',' Protective-serv',' Machine-op-inspct')
AND hours-per-week > 36.5
AND 8.5 < education-num <= 9.5
AND capital-gain <= 5095.5
AND relationship IN (' Wife',' Husband')
THEN Income = <=50K. with probability : 100.0%

Download Shih Data Miner

Shih 's main project is focused on the development of statistical and applied mathematical algorithms for prediction models. The models are used in financial time series, credit scoring applications, risk management, medical diagnosis, decision making and classification tasks.

Shih was the opposite of facts and raw information; shih was the elegance of knowledge, the insight and skill to organize knowledge into meaningful patterns. As an artist chooses colours or light to make her pictures, a master of shih chooses textures of knowledge, various ideas, myths, abstractions, and theories to create a way of seeing the world. The aesthetics and beauty of knowledge - this was shih.

[David Zindell, The Broken God 1993]