The GESIS Panel is a probability-based mixed-mode access panel (n = 4,700) based in Germany. Similar to other panel studies, the GESIS Panel is suffering from attrition due to nonresponse, which may lead to a substantial loss in data quality. Hence, it is of uttmost importance to identify panelists that are at risk of dropping out of the panel. Once these at-risk panelists have been identified researchers can take appropriate measures to prevent panelists from dropping out. Here, we will focus on identifying panelists-at-risk using statistical learning techniques, tree-based ensemble methods in particular. Our contribution has two objectives: First, we want to extent existing approaches of predictive modeling in survey methodology by including extensive data from the panel management of the GESIS Panel. Preliminary results show varying variables importance for various statistical learning techniques. Second, we aim at a comparison of various statistical learning techniques. At this early stage, the random forest technique shows the best results in terms of positive predictive value and sensitivity.