3.3.1. Decision Trees

Decision Trees are simple yet effective supervised learning models. These models work like a decision-making flowchart, where each node represents a choice based on specific feature and each branch shows result of that choice. They are especially useful for clear visual structure and interpretability, however they can overfit on complex or noisy data.

Dua created a model combining ECG signals with facial features to identify emotions like happiness, tension, and calmness. Among other models, this study also applied the Bagged Decision Trees classifier using the Gini index criterion to classify emotions from ECG and facial features, achieving overall accuracy of 85.4%. Another study by Wirawan et al. applied Decision Trees with Convolutional Neural Networks (CNNs) to classify emotions from EEG signals and compared different baseline reduction methods on the DEAP dataset. Then, Liew et al. took it further by integrating Decision Trees with genetically optimized Fuzzy ART ensembles to improve classification on arousal and valence scales. Genetically optimized Fuzzy Adaptive Resonance Theory (ART) ensembles use a genetic algorithm to improve clusters of data patterns, where each round tweaks and selects the best clusters to help recognize complex emotions more accurately.