Emotion Classification in Digital Art using Color Features and Machine Learning

Colors have a big impact on human emotions, influencing how we interpret and respond to visual stimuli. This paper presents a study on the classification of emotions in digital art using machine learning (ML). A dataset of 10,632 images collected from DeviantArt, covering ten distinct emotions, was analyzed to categorize these emotions into positive and negative groups. Various ML algorithms, including Support Vector Machines (SVM), Decision Trees, and Random Forest, were employed to predict the emotional classification of the artworks based on extracted color features. To achieve this, we implemented a structured feature extraction process, where regular colors and achromatic color proportions were identified and mapped using a predefined color dictionary. The extracted features were then used to train and evaluate classification models. Experimental results indicate that Random Forest outperforms other models, achieving the highest accuracy in distinguishing positive and negative emotions (with accuracy = 0.76). Additionally, an analysis of achromatic colors revealed a weak but notable correlation with emotion polarity. The study findings prove the potential of color-based features in automating emotion recognition in digital art. This is important in various areas besides art, including product design, social media content creation, education, etc.