Over the past few decades, the concept of circular economy has gained a wide popularity in our society and raised both individual and public consciousness. Under the slogan of Reduce, Reuse and Recycle several companies in the waste collection market have decided to rethink the way they tackle trash collection in cities and, in general, urban areas. This thesis is written in cooperation with moltosenso, an IT-company with its main seat in Turin, Italy. The collaboration with Cidiu, the waste collection company liable for the territories outside Turin, has given birth to the project analysed in this thesis. The objective is to develop and implement statistical models that can be used to describe the territory waste production. The necessary data for our statistical models comes in the form of lagged measurements from sensors mounted on the collection vehicles. Although it is common use in statistical applications and in econometrics to rely on time series, we will extend our approach considering neural networks as well. In particular, Recurrent Neural Networks (RNNs) and Long Short Term Memory (LSTM) networks come in very handy when dealing with time series objects. The ability to predict the amount of waste produced in a neighbourhood could be used in the future to reduce the pollution caused by trucks, travelling unnecessarily around the city towards almost empty dumpsters.