Artificial Intelligence Models for Analyzing Children’s Emotional Expressions and Color Use in Digital Drawings

This study examines the relationship between preschool children’s emotional expression and colour use in digital drawings produced during an authentic classroom activity. A total of 216 children created drawings and recorded an oral narration in European Spanish. An expert identified, in real time, the emotional expression conveyed by each narration and categorized it by valence as positive, neutral, or negative. To examine the consistency of these labels, Machine Learning models were trained on audio recordings and combined into meta-models. When applied to the Expert dataset of 618 audio-drawing pairs, the best meta-model achieved an accuracy of 0.85 and an F1-score of 0.83, while agreement with the expert labels was moderate, with most disagreements concentrated in the neutral category. Drawings associated with positive emotional expression showed a larger coloured area and a greater number of colours than those associated with negative expression, and a larger coloured area than those associated with neutral expression. Greater use of green was also observed in positive than in negative drawings, although this pattern may partly reflect nature-related topics represented in the drawing.