Construction of a Machine Learning Prediction Model on the Impact of Aromatic Plants on Mental Health

Aromatherapy has gained attention for its potential to reduce stress and improve emotional well-being through exposure to volatile compounds from aromatic plants. However, most existing studies rely on subjective reports and lack robust, data-driven models to predict individual psychological responses. To address this challenge, we propose a multimodal machine learning framework that integrates GC-MS chemical profiles of essential oils, physiological signals (HRV, EEG), and mood assessments. A hierarchical ensemble model—combining LightGBM, BiLSTM, and Transformer components—was developed to predict emotional outcomes with both high accuracy and interpretability. Our model achieved 90% accuracy and an AUC of 0.94 in classifying significant mood improvement, outperforming traditional methods. SHAP analysis revealed key compounds, such as linalool and <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\alpha$</tex>-pinene, that influence emotional responses and interact with baseline anxiety levels. This work bridges the gap between chemical composition and psychological effects, offering a quantitative foundation for personalized aromatherapy and mental health interventions.