This document discusses the need for predictive modeling approaches in psychiatry rather than traditional hypothesis testing and inferential approaches. It argues that predictive modeling is better suited for large datasets, high-dimensional data, multiple outcomes, heterogeneous data sources, and precision medicine driven by observational data. The document reviews previous work applying predictive modeling and machine learning techniques to brain imaging and other data to predict cognitive domains in schizophrenia, position disorders along continuous axes, and integrate diverse data types for deeper phenotyping in psychiatry.