Depression is highly prevalent, and involves many different symptom profiles, courses and prognoses. Over the years, research has consistently shown that different types of evidence-based treatments are equally effective on average. However, individual responses to these treatments vary widely and are highly unpredictable. Since it is unclear what works best for a specific individual, in clinical practice, consecutive treatments are often being offered to find the optimal regimen. This strategy can be best described as a “trial and error approach”. As alternatives for this approach, this thesis focusses on data-driven personalized treatment strategies that aim to find the optimal depression treatment for a specific individual at a specific point in time. Results of this thesis indicate that prediction models are superior to traditional approaches (e.g., clinical judgment), and have the potential to guide personalized treatment strategies in depression.