A spatially based machine learning algorithm for potential mapping of the hearing senses in an urban environment

Abstract

• Four modes of hearing senses (normal, annoying, stressful, and pleasant) were examined in an urban environment. • The relationship between people's perceived hearing senses and spatial criteria was examined. • Spatial modeling and potential mapping of hearing senses were performed using a random forest (RF) algorithm. • The results showed that the RF algorithm had good accuracy in the potential mapping of hearing senses. Mapping individuals’ sense of hearing in the urban environment helps urban managers and planners accomplish goals such as creating a favorable urban environment for the citizens. The present study has been conducted to address the lack of modeling and compilation of a sense of hearing potential map in the urban environment and can help urban managers and planners make decisions in this regard. The present study aims for spatial modeling of people's hearing senses and developing potential maps for various hearing states in Tehran, Iran, using a random forest (RF) machine learning algorithm. The four various states, including pleasant sound, annoying sound, normal sound, and stressful sound, have been considered in the present study for the sounds that can be heard in the city. First, a spatial database made up of dependent, and independent data was built. Dependent data included people's four states of hearing in the urban environment, and the respective data was collected through a questionnaire from 657 people. Independent data were categorized into four groups. The first group was traffic related noises including the criteria of traffic volume and equivalent continuous sound level (Leq), the second group included land use related criteria of distance to cemetery, distance to sports areas, distance to commercial areas, distance to primary streets, distance to secondary streets, distance to park, and distance to industrial areas. Furthermore, the public facilities related group includes the distance to airports and distance to public transportation stations, and the population related group includes population density. 70% of the data were used for training, and 30% were set aside for validation. The spatial database was used to develop the potential map for the four states of the hearing sense in the urban environment using the RF algorithm. The potential hearing sense map was evaluated using the receiver operating characteristic (ROC) curve and the respective area under the curve (AUC). The AUC values were 0.930, 0.957, 0.950, and 0.903 for each of the pleasant, annoying, normal, and stressful states, respectively. There was a higher potential for pleasant sounds in the northern districts and some of the central ones, for annoying sounds mainly in the central districts, for normal sounds in central and southern districts, and stressful sounds in some parts of the central and mainly southern districts.