Project Alien Worker
Every year, thousands of companies submit H1B visa applications to the Department of Labor on behalf of foreign workers the companies are trying to hire. These applications differ in many ways: the type of job being applied for, the company sponsoring the application, the location where the foreign worker will be based, educational qualification requirements for the job, how much the job pays, etc.
The eventual outcome of each application varies—from Rejected, Denied, Withdrawn, or Certified—the latter being what applicants (both the individuals and the companies sponsoring them) are hoping to get. Currently there’s no way for knowing how long these applications will be in "Pending" state.
For our Capstone project, we are trying to answer this basic question: Using the H1B historical data (available for 2001-2015 years), is it possible to predict how long an application for a given job type might take?
Answering this important question would serve the following purposes:
- Companies and individuals would have an idea about how long they can expect their application to take. Companies can make the necessary planning about their headcount, for instance deciding on whether to spend more time looking for, attracting and retaining closely qualified Americans, or putting this application time into consideration when making business decisions about feature deliverables and expected ship dates.
- Individuals can use this knowledge about processing time per job title to make early plans such as what job fields to get into (e.g. while in college) rather than getting into a field that customarily has long H1B visa wait times.
- Legislators can use this information to determine whether special attention should be given to some job fields whose processing times is exceptionally long.
In order to answer this question for our project, we calculated the wait times for each application, and after cleaning out any outliers (such as those with clearly wrong dates, as a result of possible transcription errors with the faxed-in applications) we used a KNN machine learning algorithm to train a model to predict the wait times with the job title as a predictor. Companies or individuals can then input a job title, and we'll use our algorithm to give them an estimated wait time before the visa is certified.
There are sites out there that touch on the fact that it takes a while for a visa to be certified. Unfortunately, none of these sites care to give predictions about how long the application might take. Furthermore, the official US immigration (USCIS) website only lets users see what the current stage of their application is—it does not tell them how long it might take. Our project therefore adds this missing prediction component to this aspect of H1B visa application, something that is currently lacking officially or otherwise.