Datasets and Benchmarks

This chapter examines the empirical foundations of EEG-based affective computing: the datasets used to train and evaluate models, and the benchmarks used to compare methods. The central question is not only how much data is available, but what kinds of task definitions, generalization claims, and evaluation protocols that data can support.

In affective EEG, benchmark design strongly influences apparent model performance. Elicitation paradigm, annotation strategy, subject diversity, session coverage, and recording realism all affect what can be learned and what can be concluded. This chapter therefore treats datasets as methodological objects rather than mere collections of files.

The goal is to help the reader judge whether a benchmark is appropriate for offline recognition, online tracking, subject-independent generalization, personalized modeling, or longitudinal robustness.