Problem Formulation

This chapter defines what it means to formulate an EEG-based affective computing problem rigorously. In this book, problem formulation goes beyond naming a label space. It also includes the prediction target, the temporal assumptions, the unit of inference, the deployment scenario, and the kind of generalization that the model is expected to achieve.

For affective EEG, this step is especially important because superficially similar studies may in fact solve very different tasks. A clip-level offline classifier, a causal affect tracker, a within-subject personalized model, and a subject-independent population model are not interchangeable. Their assumptions, difficulty, and claims should be separated clearly.

This chapter separates three decisions that are often mixed together: the predictive task, the required generalization setting, and the generative objective. It distinguishes classification, regression, state tracking, and forecasting; subject-dependent and subject-independent evaluation; within-subject missions such as cross-trial, cross-session, and historical-to-future prediction; and generation tasks such as conditional synthesis, denoising, and data augmentation. These distinctions determine how later chapters should design splits, metrics, validation protocols, and models.