Integrating Explainable AI, Deep Learning, and Causal Inference to Unveil Salinity Stress Responses in Lavandula Dentata

Abstract

This chapter explores the salinity stress response of Lavandula dentata under monoculture and co-cultivation with halophytes (Atriplex prostrata, Plantago macrorhiza) using an integrative approach combining explainable artificial intelligence, deep learning, and causal inference. Dimensionality reduction (PCA, t-SNE, UMAP) and clustering algorithms (K-Means, DBSCAN) identified stress- and culture-dependent phenotypic patterns, while supervised learning models (Random Forest, MLP) predicted cultivation conditions with high accuracy, highlighting the key role of root volume, proline, and water content. SHAP values offered model interpretability, and causal inference quantified the direct effects of co-cultivation on biomass. Results revealed that co-cultivation mitigates certain stress traits but may reduce biomass, suggesting trade-offs in physiological adaptation. This chapter demonstrates how data-driven frameworks can support agroecological strategies for salt-affected environments under climate change.