A machine learning for environmental noise classification in smart cities

Abstract

The sound at the same decibel (dB) level may be perceived either as annoying noise or as pleasant music. Therefore, it is necessary to go beyond the state-of-the-art approaches that measure only the dB level and also identify the type of the sound especially when the sound is recorded using a microphone. This paper presented a case study that considers the ability of machine learning models to identify sources of environmental noise in urban areas and compares the sound levels with the recommended levels by the World Health Organization (WHO). The approach was evaluated with a dataset of 44 sound samples grouped in four sound classes that are highway, railway, lawnmowers, and birds. We used mel-frequency cepstral coefficients for feature extraction and supervised algorithms that are Support vector machine (SVM), k-nearest neighbors (KNN), bootstrap aggregation (Bagging), and random forest (RF) for noise classification. We evaluated performance of the four algorithms to determine the best one for the classification of sound samples in the data set under consideration. The findings showed that the noise classification accuracy is in the range of 95%-100%. Furthermore, all the captured data exceeded the recommended levels by WHO which can cause adverse health effects.