In the context of increasing pressure on water bodies, many fish-based indices have been developed to evaluate the ecological status of rivers. However, most of these indices suffer from several limitations, which hamper the capacity of water managers to select the most appropriate measures of restoration. Those limitations include: (i) being dependent on reference conditions, (ii) not satisfactorily handling complex and non-linear biological responses to pressure gradients, and (iii) being unable to identify specific risks of stream degradation in a multi-pressure context. To tackle those issues, we developed a diagnosis-based approach using Random Forest models to predict the impairment probabilities of river fish communities by 28 pressure categories (chemical, hydromorphological and biological). In addition, the database includes the abundances of 72 fish species collected from 1527 sites in France, sampled between 2005 and 2015; and fish taxonomic and biological information. Twenty random forest models provided at least good performances when evaluating impairment probabilities of fish communities by those pressures. The best performing models indicated that fish communities were impacted, on average, by 7.34 ± 0.03 abiotic pressure categories (mean ± SE), and that hydromorphological alterations (5.27 ± 0.02) were more often detected than chemical ones (2.06 ± 0.02). These models showed that alterations in longitudinal continuity, and contaminations by Polycyclic Aromatic Hydrocarbons were respectively the most frequent hydromorphological and chemical pressure categories in French rivers. This approach has also efficiently detected the functional impact of invasive alien species. Identifying and ranking the impacts of multiple anthropogenic pressures that trigger functional shifts in river biological communities is essential for managers to prioritize actions and to implement appropriate restoration programmes. Actually implemented in an R package, this approach has the capacity to detect a variety of impairments, resulting in an efficient assessment of ecological risks across various spatial and temporal scales.