Research on Phubbing has received a lot of attention in recent years from the research community. However, the studies conducted are mainly based on linear statistics, which is a very conservative method for data analysis. To overcome this limitation, we adopted a data mining and machine learning-based approach to identify the patterns related to Phubbing behavior. We developed several models on online survey data that we collected for our analysis purposes. The results highlighted that addiction measures fail to predict Phubbing fully. Indeed, Phubbing appeared to be linked in a non-linear way to both Information and Communication Technology (ICT) measures that do not imply a dysfunctional use of technology and social anxiety. Moreover, the machine learning approach appeared more suitable than traditional linear statistics methods to predict Phubbing, as highlighted by a much higher explained variance. Phubbing is not solely attributable to addiction dynamics. Phubbing is indicated by a series of predictors that cannot be reduced to addiction (e.g., age, social anxiety, ICT services owned). Modelling procedures able to account for non-linearity are also required to accurately assessing users’ Phubbing levels. The patterns produced by our modelling procedure may help scholars in accounting for phubbing definition, detection, and prediction more accurately.