Machine learning model provides stress biomarkers for the classification of abiotic stress in Micro-Tom.

Water deficit, salinity, and cadmium (Cd) contamination have generated an environmental problem worldwide, leading to damages to plant growth due to alteration in their metabolism. This study aimed to classify enzymatic and non-enzymatic antioxidant systems in Micro-Tom (MT) plants when subjected to two intensities (moderate and severe) of water deficit, salinity, and Cd exposure. The experimental design was a completely randomized 3 × 2 factorial, with the first factor representing the stress agents (water deficit, salinity, and Cd) and the second factor indicating stress intensities (moderate and severe), along with a control group. After an acclimation period, plants were exposed to 10 days of stress. Water deficit treatments were imposed using solutions adjusted to osmotic potentials of − 0.40 MPa and − 1.00 MPa; salinity stress was established with nutrient solutions containing 40 mM or 120 mM NaCl; and Cd stress was induced using nutrient solutions with 0.25 mM or 0.5 mM CdCl₂. Laboratory analyses included lipid peroxidation, hydrogen peroxide content, proline accumulation, protein quantification, and enzyme extraction. Descriptive analyses, and a Spearman’s correlation, identified the behavior of enzymatic and non-enzymatic systems for each stress agents and intensities, enabling the selection of key influencing factors. A factorial analysis of variance was performed to assess the mean differences among the treatments (α = 0.05) for enzymatic, non-enzymatic systems, MDA and, H₂O₂. Using this data, a decision tree model classified the stresses into four levels: low, low-medium, medium-high, and high. Variations in antioxidant response and stress biomarkers were detailed, with proline and superoxide dismutase identified as the primary variables of significance across stress indicators. Furthermore, the model achieved robust classification performance with Matthew’s correlation coefficients exceeding 0.80 in the extreme classes; however, it encountered limitations in distinguishing between classes with closely proximate values. The findings indicate the capability of the decision tree to classify stress levels in plants.