Explainable AI for Urban Land Cover Classification Using Mobile Application Traffic Data

Abstract

This research explores the use of mobile application traffic data to interpret urban land cover classification using explainable machine learning methods. The experiments using a high-resolution mobile service traffic data in Paris, France show that the hourly downlink traffic of Microsoft Office, Netflix, and Uber together with the XGBoost model can accurately classify land cover types and the SHAP values help interpret instance-level feature importance and their spatial patterns.