As cloud systems continue to advance and grow in complexity, the process of manual troubleshooting becomes tedious. Instances of cloud failures can result in substantial costs for both users and cloud providers. By adopting a data-driven approach for predicting the cloud system health, Ericsson can improve system availability. Logs record noteworthy system states and events as they occur, offering valuable information for system monitor- ing. In this thesis, a data-driven method is proposed to predict the cloud system health using clustered logs. The proposed method is able to predict failures with a macro F1 score of 0.995 ± 0.004 on the train dataset and 0.982 on the test dataset. It is concluded that clus- tered logs show promising potential as a source for predicting the cloud system health.