Replicable AI framework for emotion recognition validated against human

A team from five universities in Ecuador and Spain says emotion recognition AI can be made more transparent—and therefore easier to scrutinize—without relying only on more data or deeper neural networks. In a study published in the Journal of Ambient Intelligence and Humanized Computing, the researchers proposed a replicable Human Emotion Recognition (HER) framework that combines classical machine-learning classifiers, symbolic rule mining, and validation by human experts at each stage. Instead of chasing benchmark accuracy alone, the work tests whether AI judgments are logically consistent with how trained observers reason. The approach uses Sequential Minimal Optimization, J48 decision trees, Multilayer Perceptrons, and Apriori association rules.