Predict if salary is >50k with Keras (machine learning)

A neural network with two layers (sigmoid activation) is used to predict if someone makes more than 50k or less based on training and test data from the UCI Machine Learning Repository.

Percentage right: 85.333% on 16281 test entries on which we did not trained the neural network. Therefore, the error is of 14.667%.

Dependencies

• Python 3.0 (Anaconda distribution is preferred),
• Theano (requires linux),
• Keras (requires Theano),
• iPython notebook (included in Anaconda bundle, run ipython notebook in terminal from the project's directory).

Comparison with other regression methods on the dataset

Algorithm	Error (%)
My neural network	14.67
C4.5	15.54
C4.5-auto	14.46
C4.5 rules	14.94
Voted ID3 (0.6)	15.64
Voted ID3 (0.8)	16.47
T2	16.84
1R	19.54
NBTree	14.10
CN2	16.00
HOODG	14.82
FSS Naive Bayes	14.05
IDTM (Decision table)	14.46
Naive-Bayes	16.12
Nearest-neighbor (1)	21.42
Nearest-neighbor (3)	20.35
OC1	15.04
Pebls	Crashed.

Dataset's structure

2 Outputs (classification)

• >50K,
• <=50K.

14 Inputs (string and continuous)

• age: continuous.
• workclass: Private, Self-emp-not-inc, Self-emp-inc, Federal-gov, Local-gov, State-gov, Without-pay, Never-worked.
• fnlwgt: continuous.
• education: Bachelors, Some-college, 11th, HS-grad, Prof-school, Assoc-acdm, Assoc-voc, 9th, 7th-8th, 12th, Masters, 1st-4th, 10th, Doctorate, 5th-6th, Preschool.
• education-num: continuous.
• marital-status: Married-civ-spouse, Divorced, Never-married, Separated, Widowed, Married-spouse-absent, Married-AF-spouse.
• occupation: Tech-support, Craft-repair, Other-service, Sales, Exec-managerial, Prof-specialty, Handlers-cleaners, Machine-op-inspct, Adm-clerical, Farming-fishing, Transport-moving, Priv-house-serv, Protective-serv, Armed-Forces.
• relationship: Wife, Own-child, Husband, Not-in-family, Other-relative, Unmarried.
• race: White, Asian-Pac-Islander, Amer-Indian-Eskimo, Other, Black.
• sex: Female, Male.
• capital-gain: continuous.
• capital-loss: continuous.
• hours-per-week: continuous.
• native-country: United-States, Cambodia, England, Puerto-Rico, Canada, Germany, Outlying-US(Guam-USVI-etc), India, Japan, Greece, South, China, Cuba, Iran, Honduras, Philippines, Italy, Poland, Jamaica, Vietnam, Mexico, Portugal, Ireland, France, Dominican-Republic, Laos, Ecuador, Taiwan, Haiti, Columbia, Hungary, Guatemala, Nicaragua, Scotland, Thailand, Yugoslavia, El-Salvador, Trinadad&Tobago, Peru, Hong, Holand-Netherlands.

Note: some data contain "?", so this value is possible for each of those 14 input fields

Presentation's slides

The slides of the seminar are available here (in French).

References

Dataset infos

Source:

Lichman, M. (2013). UCI Machine Learning Repository [http://archive.ics.uci.edu/ml]. Irvine, CA: University of California, School of Information and Computer Science.