Electronic noses used for outdoor ambient air characterization to assess odor impacts on population can produce large datasets since usually the sampling is conducted with high frequency (e.g. data per minute) for periods that can reach several months, with a number of sensors that ranges usually from four-six as a minimum, up to above thirty. The environmental analyst has thus to deal with large datasets (millions of data) that have to be properly elaborated for obtaining meaningful interpretation of the instrumental signals. A recent review questioned the capability of some classic statistical elaboration tools for application to e-noses, highlighting how very few in field application are present in scientific literature. In the present work we describe: (i) the use of Self-Organizing Map (SOM) algorithm as a tool for analysis and visualization of e-nose raw data collected at a receptor site near a bio-waste composting facility; (ii) a second level clusterization using k-means clustering algorithm to identify "air types" that can be detected at the receptor and (iii) the use of e-nose data related to the plant odour sources as well as odour measurements of ambient air collected at the receptor site, to classify the air types. Eventually we evaluate the frequency and duration of the air type/s identified as malodorous.