ABSTRACT
We introduce supervised, unsupervised, reinforcement, and deep learning for social data, focusing on use cases such as outcome prediction, population subtyping, image/text understanding, and simulation-based policy search. Practical sections cover data preparation (integration, imputation, scaling, and feature engineering), pipeline reproducibility, cross-validation, and drift monitoring. We stress model interpretability (global/local explanations), bias diagnostics across subgroups, and theory integration to elevate findings from correlation to social mechanisms. Case vignettes show how interpretable ML uncovers multifactor interactions (e.g., inequality moderators), how reinforcement learning (RL) can explore policy spaces in silico, and how transfer learning “lowers the data bar” for domain studies. The chapter closes with “best-practice checklists” for robust, ethical, and theory-aware ML in social research.
