Odor-causing substances are generated by various emission sources in urban areas. Recently, urbanization has greatly increased the density of odor emission facilities, implying the identification of odorants emission source is challenging. Identifying emission source is multifactorial, and a machine learning approach is considered useful for these complicated matters. The objectives of this study were to propose a method using machine learning-based classification models to identify odor sources in urban areas. We collected 34,539 data points regarding quantitative data of 22 compounds emitting from 11 types of facilities in urban areas (i.e., automobile industry, bio factory, wastewater treatment plant, landfill, construction site, farm industrial complex area, restaurant, gas station, roadside, park) and odor intensity of these 11 facilities. Decision tree (DT) and random forest (RF) algorithms were used as classification models for identifying odor sources with 23 variables (22 compounds + odor intensity). The DT model identified 7 out of 11 emission sources with 87.15% accuracy. The RF model identified all 11 emission sources with 99.23% accuracy. When including 6 important variables only (i.e., hydrogen sulfide, ammonia, trimethylamine, methyl mercaptan, acetaldehyde, odor intensity) in the RF model, accuracy (99.15%) was almost same with that (99.23%) obtained from all 23 variables included as variables in the model. Our findings imply that a machine learning approach can help to identify odor emission sources with high accuracy and we can save time and cost in the identification of odor emission sources by including the 6 important variables only. • Identified odor emission sources in urban areas using machine learning algorithms. • Collected 34,539 data sets of 22 substances from 11 types of emission sources. • Random forest model identified all 11 emission sources with 99.23% accuracy. • Simplified RF model with 6 important variables presented the almost same accuracy.