A general procedure to generate models for urban environmental-noise pollution using feature selection and machine...

Area of Science:

• Environmental Science
• Acoustics
• Computer Science

Background:

• Urban noise prediction is a complex, non-linear problem due to numerous interacting variables and spatial heterogeneity.
• Accurate modeling of environmental noise in cities is essential for urban planning and public health.

Purpose of the Study:

• To develop and apply a procedure for accurate environmental noise prediction in urban areas.
• To evaluate the effectiveness of feature selection techniques and machine learning regression methods for predicting the energy-equivalent sound-pressure level (LAeq).

Main Methods:

• Employed three machine learning regression methods: multilayer perceptron (MLP), sequential minimal optimisation (SMO), and Gaussian processes for regression (GPR).
• Utilized three feature selection/data reduction techniques: correlation-based feature-subset selection (CFS), wrapper for feature-subset selection (WFS), and principal component analysis (PCA).

Main Results:

• The combination of WFS with SMO or GPR yielded the most accurate LAeq estimations, achieving an R-squared value of 0.94 and a mean absolute error (MAE) between 1.14 and 1.16 dB(A).
• Different schemes were proposed based on varying data collection and accuracy requirements.

Conclusions:

• Feature selection, particularly WFS, is crucial for simplifying complex urban noise models.
• Machine learning regression methods, especially SMO and GPR, are highly effective for non-linear environmental noise prediction, offering practical solutions for urban acoustic management.