From description to code: a method to predict maintenance codes from maintainer descriptions

Aircraft maintenance crews enter the actions performed, time required to complete actions, and process followed action into a system of record that may be used support future important operational decisions such as part inventory staffing levels. Unfortunately, performed by maintainers not align with structured, predetermined codes for actions. This discrepancy combined an overabundance structured has led incorrect polluted data cannot in decision making. Typically, unstructured textual fields accurately action, but are inaccessible common reporting approaches. The can cleanse fields, thereby making more available paper introduces natural language processing pipeline predict C-17 US Air Force from unstructured, shorthand text record. research aims problematic further use efficiency asset reliability measures. Novel processing, extraction, clustering, classification approaches was employed develop suited peculiarities short, jargon-based text. evaluates frequency field values within datase selects appropriate machine learning model optimize predictive accuracy. Three different methods were investigated determine optimal approach: Logistic Regression Classifier, Random Forrest Unsupervised techniques. predicted average accuracy 93 % across five codes.