Lean blow-out (LBO) is a critical phenomenon in gas turbines. It is enhanced by very to ultra-lean operating conditions which are considered today to decrease the environmental impact of combustion. Despite many studies on the subject, the physical mechanisms leading to global flame extinction are not fully understood. Recently, unsupervised classification has appeared in the literature as an efficient tool to identify key features in reactive flows. In this work, unsupervised classification relying on Principle Component Analysis (PCA) and K-means clustering algorithms is used to investigate the underlying mechanisms of a blowoff event in a bluff body configuration. The unsupervised classification allows to identify and track in time 4 distinct zones: fresh gases, burnt gases, fast reacting flame and preheat zone. To elucidate the blow-out processes, an analysis of mass and energy balances of the different zones is proposed. This analysis describes the temporal evolution of the zones as a result of their interactions, which is the main driver for flame stabilization or blowoff. For the considered blowoff event, the extinction is induced by an imbalance between the various contributions identified by the proposed analysis: while the decrease in fuel mass flow rate modifies both the conductive fluxes and chemistry source terms in the reactive zones, the convective fluxes remain constant over time as the total mass flow rate is kept constant. This work suggests that the proposed methodology is a useful tool to analyze unsteady configurations and understand the main mechanisms at work in such configurations.