The goal of this project is to find a model that accurately predicts if a person is divorced, single or married. We think that there are a series of social, economic, racial, religious and geographic variables that relate to a certain person being divorced.
We are ceating a tool that accurate predicts if your are marriage status. Wether you are divorced, single or married, this tool will predict with an accuracy in the 80s percentage, given certain characteristics.
We use decision trees model to predict the outcomes (divorced, single or married). We create a prediction with CP = 0 (most precise and really high computing power). After it, we create a loop and start "pruning" the tree with it, in order to get the highest mean F1 and CP possible. As it turns out, our mean F1 on the the test sample is around 81% with a CP = 2.176e^(-5).
We use the Current Population Survey (http://thedataweb.rm.census.gov/ftp/cps_ftp.html). We downloaded the data using a parsing script written within DOC and given by Prof Jeff Chen for public use. We modified it to get the data for the twelve months of 2016.
We have approximately 1.8 million observations and 389 variables. From those 389 variables available in the dataset we use 35 variables to predict the outcome. These are the variables used with their importance:
| varname | importance | DESCRIPTION |
|---|---|---|
| prtage | 48027.2043 | PERSONS AGE |
| hrnumhou | 19580.93662 | TOTAL NUMBER OF PERSONS LIVING |
| prnmchld | 10172.78591 | Number of own children < 18 years of age |
| prchld | 10130.38892 | PRESENCE OF OWN CHILDREN < 18 YEARS |
| pemlr | 4419.647807 | MONTHLY LABOR FORCE RECODE |
| hefaminc | 4143.967869 | FAMILY INCOME |
| puwk | 3366.378441 | LAST WEEK, DID YOU DO ANY WORK |
| prtfage | 2525.231903 | TOP CODE FLAG FOR AGE |
| peeduca | 2101.682639 | HIGHEST LEVEL OF SCHOOL |
| prwksch | 1873.531408 | LABOR FORCE BY TIME |
| prcivlf | 1733.776621 | CIVILIAN LABOR FORCE |
| ptdtrace | 1521.411985 | RACE |
| prinusyr | 1411.366186 | IMMIGRANT'S YEAR OF ENTRY |
| prwkstat | 1346.103785 | FULL/PART-TIME WORK STATUS |
| prmjind1 | 762.5797876 | MAJOR INDUSTRY RECODE - JOB 1 |
| gediv | 726.3161318 | DIVISION |
| gtcbsasz | 656.6140444 | Metropolitan Area (CBSA) SIZE |
| gtindvpc | 554.5695261 | INDIVIDUAL PRINCIPAL CITY |
| prcitshp | 543.2361561 | CITIZENSHIP STATUS |
| prmjocc1 | 533.5237143 | MAJOR OCCUPATION RECODE |
| pesex | 530.7090073 | SEX |
| gereg | 389.8060756 | REGION |
| gtcbsast | 383.5875216 | PRINCIPAL CITY/BALANCE STATUS |
| hehousut | 268.7665871 | TYPE OF HOUSING UNIT |
| hurespli | 256.9704462 | LINE NUMBER OF THE CURRENT |
| gtmetsta | 183.9798964 | METROPOLITAN STATUS |
| peafever | 181.6559453 | DID YOU EVER SERVE ON ACTIVE |
| huspnish | 125.2598896 | IS SPANISH THE ONLY LANGUAGE SPOKEN |
| peafnow | 109.5453365 | ARE YOU NOW IN THE ARMED FORCES |
| penlfact | 106.0165516 | WHAT BEST DESCRIBES YOUR SITUATION AT |
| pehspnon | 85.31919456 | HISPANIC OR NON-HISPANIC |
| pedisrem | 65.85950161 | BECAUSE OF A PHYSICAL, MENTAL, OR |
| hubus | 49.12820317 | DOES ANYONE IN THIS HOUSEHOLD |
| pedisear | 17.44051162 | IS...DEAF OR DOES...HAVE SERIOUS DIFFICULTY HEARING? |
| prftlf | 9.76262516 | FULL TIME LABOR FORCE |
The parsing script also creates a dictonary for these variables.
Our project can be checked in the divorce-prediction.Rmd file.
This file project was broken down into 7 main parts
- 1 Installing all the libraries needed.
- 2 Declaring functions that will be used in other parts of the project.
- 3 Creating a Data Dictionary.
- 4 Downloading the data.
- 5 Culling and cleaning the data.
- 6 Training the model.
- 7 Optimizing the model.
- 8 Iterating to find the best mean F1.
NOTE: in order for this script to work working directory and paths in the files must be updated to fit the user
- April 17, 2017: We started wrinting our read-me file and gathering the data.
- April 20, 2017: Updates on Read-me file.
- April 24, 2017: Finishing Read-me file.
- April 30, 2017: Finishing the code.
- May 8, 2017: Code Finished and Read-Me Finished.
Credits of this work go out to Carlos Delgado and Christian Ueland.
No license needed.