Machine learning in the prediction of binge drinking: A systematic review

Binge drinking (BD) refers to episodes of excessive alcohol consumption, which are associated with various acute and long-term consequences. Despite its high prevalence and negative impacts, predictive features for BD remain lacking. This study aims to review machine learning (ML) research on BD, comparing different algorithms and types of data, trying to identify promising models and variables that may help identify individuals at risk.

A systematic search was performed in PubMed, Embase, and Scopus to identify relevant studies published before July 2, 2026. 21 studies were included in data extraction process.

The results suggest that impulsive factors, substance-use variables, emotional distress, as well as demographic features, particularly gender and age, play a significant role in predicting BD. Moreover, emerging methods that use real-time data recorded via mobile apps showed promise. Conversely, findings from neural data yielded mixed results, with no single brain area consistently associated with BD. The qualitative comparison did not show any specific ML algorithm consistently outperforming the others across studies; however, the predictive performance of ML approaches tended to improve when different data types were combined.

The analysis of the available literature was restricted due to the non-homogeneous characteristics of studies, mainly in terms of methodology, which did not allow us to draw comprehensive conclusions.

The comparison of available ML studies showed how promising risk factors can be combined within multivariate models to increase their predictive value. Future large-scale studies using multisource data may better define risk profiles to foster prevention strategies.