An auto-encoder based LSTM model for prediction of ambient noise levels. (30th June 2022)

• Record Type: Journal Article
• Authors: Tiwari, S.K.; Kumaraswamidhas, L.A.; Gautam, C.; Garg, N.
• Is Part Of: Applied acoustics. Volume 195(2022)
• Publication Date: 2022-06-30
• Subjects: Deep learning -- Auto-encoder -- LSTM -- Data preprocessing -- Noise level prediction -- Friedman test; Acoustical engineering -- Periodicals; Periodicals; 620.2
• DOI: 10.1016/j.apacoust.2022.108849
• Languages: English
• ISSNs: 0003-682X

Abstract

Highlights: An auto-encoder based LSTM model is proposed for ambient noise levels predictions and forecasting. The proposed model outperfroms the conventional time-series modelling techniques. The performance of the model is ascertained by some statistical tests. Abstract: Traffic noise is one of the most prevalent cause of environmental pollution in Indian cities. A reliable method is required for assessment, and prediction of ambient noise levels. This paper presents a novel deep learning model based on Auto-encoder infused with Long short-term memory (LSTM), to predict ambient noise levels. The model automatically selects the best prediction technique by considering different combination of hyper-parameters using grid search methodology. It has the ability to inherit non-stationary characteristics of time-series data while considering non-linear pattern. The proposed model is compared with some well-known techniques like Artificial neural technique (ANN), Support vector machine (SVM), Recurrent neural network (RNN), and Long short term memory (LSTM) model. The study concludes that the proposed model outperforms other techniques and can be a reliable approach for time-series prediction of ambient noise levels with an error of ± 0.563 dB(A). The prediction capability of the models is ascertained by statistical tests parameters namely RMSE, MAE, R 2, and ACC % which is further validated by Friedman test.