The entrainment of material is a common process in debris-flow behaviour and can strongly increase its total volume. However, due to the complex nature of the process, the exact mechanisms of entrainment have not yet been solved. We analysed geomorphological and topographical data collected in 110 reaches of 17 granular debris flows occurred in the Pyrenees and the European Alps. Four governing factors sediment availability, channel-bed slope, channel cross section shape and upstream-contributing area) were selected and defined for all the 110 reaches. One dataset of the resulting database was used to develop two models to estimate the erosion rates based on the governing factors: a formula derived from multiple linear regression (MLR) analysis and a decision tree (DT) obtained from J48 algorithm. The models obtained using these learning techniques were validated in another independent dataset. In this validation set, the DT model revealed better results. The models were also implemented in a torrent (test set), where the total debris-flow volume was known and two empirical methods (available in literature) were applied. This test revealed that both MLR and DT predict more accurately the final volume of the event than the empirical equations for volume prediction. Finally, a general DT was proposed, which includes three governing factors: sediment availability, channel-bed slope and channel cross section shape. This DT may be applied to other regions after adapting it regarding site-specific characteristics

Peer Reviewed