As machine learning techniques increase in accuracy and availability, their potential to automate tedious, time-consuming tasks grows. When applied across a multi-million-member federal workforce, high quality results can produce dramatic savings. One task ripe for machine learning is the classification of job grade for federal employees. To date, human resource staff and hiring managers have manually assigned a grade to indicate the seniority of each position, a monumental burden in person-hours with inherent inconsistency - and therefore potential for bias and inequity among employees.

A 92%-accurate model was developed using natural language processing (NLP) vectorization, transfer learning, and tree regression to assign job grade based on the text in a job posting. This application of deep learning boosts efficiency in human resource management through direct reduction of time spent on repetitive tasks and improves equity among employees by highlighting disparities in the historical human-assigned grades. Artificial intelligence can reduce this burden and increase standardization using natural language processing (NLP) techniques.

The thesis paper, co-authored by Jordan Jasuta Fischer, Justin A Pavis, Keston M Crandall and Justin Saunders can be found here.