An Airport Noise Prediction Model Based on Selective Ensemble of LOF-FSVR

Abstract

Airport noise prediction is of vital importance to the planning and designing of airports and flights, as well as in controlling airport noise. Benefited from the development of the Internet of things, large-scale noise monitoring systems have been developed and applied to monitor the airport noise, thus a large amount of real-time noise data has been collected, making it possible to train an airport noise prediction model using an appropriate machine learning algorithm. Support vector machine (SVM) is a powerful machine learning algorithm and has been demonstrated to have better performance than many existing algorithms. Thus, we intend to adopt SVM as the base learning algorithm. However, in some cases, the monitored airport noise data contains many outliers, which degrades the prediction performance of the trained SVM. To enhance its outlier immunity, in this paper, we design a Local Outlier Factor based Fuzzy Support Vector Regression algorithm (LOF-FSVR) for airport noise prediction, in which we calculate the fuzzy membership of each sample based on a local outlier factor. In addition, ensemble learning has become a powerful paradigm to effectively improve the prediction performance of individual models, motivated by ensemble learning, we propose a LOF-FSVR based ensemble prediction model to improve the prediction accuracy and reliability of single LOF-FSVR airport noise prediction model. Conducted experiments on the monitored airport noise data demonstrate the good performance of the proposed airport noise prediction model.