Machine Learning Assisted Evaluation of Gibberellic Acid (GA₃) Foliar Application on Raphanus sativus L Under PotassiumPermanganate Induced Oxidative Stress: A Data-Driven Agronomic and Biochemical Analysis

Abstract

Soil salinity and salt linked oxidative stress remain among the most limiting abiotic constraints on vegetable productivity in the arid and semi-arid belts of Khyber Pakhtunkhwa, Pakistan, and radish (Raphanus sativus L.) is particularly sensitive to such conditions. This study evaluates the ameliorative role of exogenously applied gibberellic acid (GA₃) on the agronomic and biochemical performance of R. sativus subjected to graded potassium permanganate (KMnO₄)–induced oxidative stress, and couples the classical wet-lab experiment with a lightweight, reproducible machine-learning analytics layer. A randomized complete block design was used with four KMnO₄ levels (0, 20, 30, 40 µM) and GA₃ doses ranging from 75 to 200 ppm. Growth attributes and biochemical markers chlorophyll a/b, carotenoids, soluble sugars, protein and total phenolics were digitized and processed through an automated pipeline combining analysis of variance, Pearson correlation mapping, random-forest feature attribution and gradient-boosted regression under leave-one-out cross-validation. KMnO₄ stress alone sharply depressed every measured trait, whereas GA₃ supplementation produced a clear dose-dependent recovery; the 200 ppm GA₃ + 40 µM KMnO₄ combination restored chlorophyll a, chlorophyll b and carotenoid pools and lifted protein and phenolic content well above the stressed control. The data-driven models confirmed GA₃ dose as the dominant positive predictor of both growth and biochemical status (feature importance and correlation r ≈ 0.82–0.93), while KMnO₄ acted as the principal stress driver. These findings demonstrate that GA₃ mitigates salt-linked oxidative injury likely by stabilizing membrane integrity and promoting protein and pigment biosynthesis and that interpretable machine-learning models can add quantitative, transparent value to conventional plant stress phenotyping.