AffectEval: A Modular and Customizable Framework for Affective Computing

TL;DR

AffectEval tackles the lack of reusable, cross-domain frameworks in affective computing by delivering a modular, object-oriented pipeline with six core components plus two added modules (Feature Selector and Label Generator). It demonstrates its utility by reproducing Schmidt et al. (2018) and Zhou et al. (2023) experiments on APD and WESAD with comparable or better performance and up to about 90% fewer lines of code. The framework standardizes data formats, supports multimodal physiological signals, and emphasizes easy customization, enabling rapid construction of end-to-end affective computing pipelines across domains. Its open-source release is intended to foster broader adoption and fair, reproducible comparisons in the field.

Abstract

The field of affective computing focuses on recognizing, interpreting, and responding to human emotions, and has broad applications across education, child development, and human health and wellness. However, developing affective computing pipelines remains labor-intensive due to the lack of software frameworks that support multimodal, multi-domain emotion recognition applications. This often results in redundant effort when building pipelines for different applications. While recent frameworks attempt to address these challenges, they remain limited in reducing manual effort and ensuring cross-domain generalizability. We introduce AffectEval, a modular and customizable framework to facilitate the development of affective computing pipelines while reducing the manual effort and duplicate work involved in developing such pipelines. We validate AffectEval by replicating prior affective computing experiments, and we demonstrate that our framework reduces programming effort by up to 90%, as measured by the reduction in raw lines of code.

Figures (5)

• Figure 1: Key components of an affective computing pipeline, identified in metafera, and examples of actions performed by each.
• Figure 2: Overview of AffectEval components. We extend the components identified by metafera with Feature Selection and Label Generation, which are useful and often necessary steps for affect recognition across domains. Bolded points indicate pre-implemented behaviors.
• Figure 3: Example instantiation of AffectEval. Solid outlines indicate required components, and dashed outlines indicate optional components.
• Figure 4: The workflow of AffectEval-based implementation to replicate 3-class affect recognition and binary stress detection on WESAD schmidt2018 and binary stress detection on APD and WESAD zhou2023.
• Figure 5: Comparison of manual effort required for pipeline implementation.