Multidimensional Noise Pollution Modeling Through Integrated Signal Processing and Machine Learning Techniques in Nepalese Urban Corridors

Abstract

This paper explores noise pollution trends in Makwanpur, Chitwan and Kathmandu using signal processing and machine learning to explore the statistical characteristics of noise in the environment and forecast its temporal variations. Calibrated sound level meters were then used to measure noise levels on the hubs of major roads, business centers, residential and sensitive areas according to the A-weighted decibel scale. Filtering, normalization, and time-series standardization were done to preprocess the recordings, and statistical descriptors of the mean, variance, peak levels, and percentile distributions were calculated. The Support Vector Machines and Artificial Neural Networks were then used to predict the correlation between the factors and real noise levels. The findings indicated that Kathmandu has the maximum intensity of noise, variability and the maximum exposure which are then followed by Chitwan and Makwanpur. The predictive ability of the machine learning models was shown to be high, ANN was superior to SVM in the ability to capture non-linear noise dynamics since it has lower RMSE and MAE values, and better accuracy. The results validate the assumption that the statistical indicators are strong measures of the structure of noise pattern underlying and also enhance prediction significantly. On the whole, the research concludes that machine learning is a dependable model when it comes to predicting the noise pollution and can serve as a useful resource to environmental management, urbanism planning, and safety of the population in fast-paced urbanizing districts.