Date of Award

3-22-2018

Document Type

Thesis

Degree Name

Master of Science in Operations Research

Department

Department of Operational Sciences

First Advisor

Raymond R. Hill, PhD.

Abstract

An Air Force sponsor is interested in improving an acoustic detection model by providing better estimates on how to characterize the background noise of various environments. This would inform decision makers on the probability of acoustic detection of different systems of interest given different levels of noise. Data mining and statistical learning techniques are applied to a National Park Service acoustic summary data set to find overall trends over varying environments. Linear regression, conditional inference trees, and random forest techniques are discussed. Findings indicate only sixteen geospatial variables at different resolutions are necessary to characterize the first ten ⅓ octave band frequencies of the L90 band using just the linear regression. The accuracy of the regression model is within 2 to 6 decibels and depends on the frequency of interest. This research is the first of its kind to apply multiple linear regression and a conditional inference tree to the national park service acoustic dataset for insights on predicting noise levels with dramatically less variables than needed in random forest algorithms. Recommended next steps are to supplement the national park service dataset with more geographic information system variables in common global databases, not unique to the United States.

AFIT Designator

AFIT-ENS-MS-18-M-155

DTIC Accession Number

AD1056413

Share

COinS