Machine Learning Meets the Subconscious: Can AI Decode Dreams?

This study examines the relationship between the structured features of dream narratives and the presence of negative emotional content. We extract and engineer features that include demographic indicators (e.g., gender), character composition (e.g., proportions of family, friends, animals, dead or imaginary figures) and emotional behavior indices (Aggression/Friendliness). To model the relationship between dream characteristics and negative emotions, we implement a classification algorithm, k-nearest neighbors (KNN) and multiple regression algorithms: linear regression, k-nearest neighbors (KNN), support vector machine (SVM), random forest, and extreme gradient boost (XGBoost), and evaluate performance using multiple evaluation metrics such as Mean Squared Error (MSE). XGBoost achieved the lowest MSE (0.2065), outperforming other models in capturing non-linear patterns. Among the features, Aggression/Friendliness demonstrate the strongest associations with negative emotions, while Friendliness and sexuality show moderate correlations. Gender has minimal impact, suggesting that dream content may be largely independent of the dreamer’s gender. Although the overall predictive performance of the models is modest, the findings provide insight into the dream elements that can influence the emotional tone. This work highlights both the promise and limitations of using machine learning for effective dream analysis and suggests future directions that integrate psychological or contextual variables for improved emotion prediction.