The Berliner Wasserbetriebe (BWB) are operating more than 650 vertical filter wells supplying the drinking water for the nearly 3.7 million inhabitants from groundwater resources within the city limits. In order to keep performance and water quality as high as possible, these wells require regular monitoring and maintenance. The main reason for inefficient well performance is so-called well ageing caused by deposit formation due to multiply correlated biological, chemical and physical clogging processes in and around the well that decrease the yield for a given drawdown. In order to better understand the key drivers for well ageing and to project the loss of capacity for a given time ahead, machine learning (ML) approaches were applied to selected data from routine well monitoring. The statistical programming language R was used for automated data processing, feature selection and assessment of the importance of the selected variables, and finally for model training and prediction of future loss of well capacity. Four variables were identified to be highly significant predictor variables. Multivariate linear regression, logistic regression, decision tree, random forest and gradient boosting were applied, the latter performing best with a sensitivity of 94% and precision of 88%. The approach is now transferred into a well condition index to be included in a well management and reporting tool box developed in the frame of the H2020 project digital-water.city.