Continuous and high acoustical noise level is one of the recurrent problems that affect citizens of large cities. For this reason, acoustic level long term monitoring is a common practice in large urban areas, in which, according to the international regulation, the noise levels must be kept under certain thresholds. Frequently, in order to predict acoustical level values in future periods, forecasting methods are implemented. Many of these techniques need a calibration or training phase to be performed on a continuous measurements dataset, i.e. not affected by missing data. In this paper the performances in the reconstruction of missing data of two techniques are compared. The models implemented are a Time Series Analysis (TSA), based on the evaluation of trend and periodicity of the series, and a Regression (REGR) method, based on a modification of linear stochastic regression. The error analysis will show interesting features of both the models. In addition, the study of dataset mean and variance preservation will highlight the differences between a deterministic (TSA) and a stochastic (REGR) imputation approach. Finally, a validation on 21 data not used in the calibration phase is presented, comparing the predictive performances of two TSA models, calibrated on datasets with 60 missing data and in which different imputation techniques have been used.

A comparison of imputation techniques in acoustic level datasets

GUARNACCIA, CLAUDIO;QUARTIERI, Joseph;TEPEDINO, CARMINE;
2015

Abstract

Continuous and high acoustical noise level is one of the recurrent problems that affect citizens of large cities. For this reason, acoustic level long term monitoring is a common practice in large urban areas, in which, according to the international regulation, the noise levels must be kept under certain thresholds. Frequently, in order to predict acoustical level values in future periods, forecasting methods are implemented. Many of these techniques need a calibration or training phase to be performed on a continuous measurements dataset, i.e. not affected by missing data. In this paper the performances in the reconstruction of missing data of two techniques are compared. The models implemented are a Time Series Analysis (TSA), based on the evaluation of trend and periodicity of the series, and a Regression (REGR) method, based on a modification of linear stochastic regression. The error analysis will show interesting features of both the models. In addition, the study of dataset mean and variance preservation will highlight the differences between a deterministic (TSA) and a stochastic (REGR) imputation approach. Finally, a validation on 21 data not used in the calibration phase is presented, comparing the predictive performances of two TSA models, calibrated on datasets with 60 missing data and in which different imputation techniques have been used.
File in questo prodotto:
Non ci sono file associati a questo prodotto.

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: http://hdl.handle.net/11386/4653066
 Attenzione

Attenzione! I dati visualizzati non sono stati sottoposti a validazione da parte dell'ateneo

Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 8
  • ???jsp.display-item.citation.isi??? ND
social impact