<p>Agricultural diseases outbreak poses a serious threat not only to crop yield but also to the stability of agroecosystems, food security and environmental sustainability. Harumanis mango (<i>Mangifera indica</i> L.), a commercially important tropical fruit cultivated widely across diverse regions, is particularly vulnerable to foliar disease that reduces production and interferes with post-harvest processes. Timely, scalable solutions are necessary since traditional disease detection methods that rely on visual inspection are frequently labor-intensive and prone to late symptom recognition. This study proposes an optimized ensemble learning framework for predicting leaf disease severity in mango crops using image-derived morphological features. Several regression models, including Support Vector Machine, Gradient Boosting, and XGBoost, were trained and dynamically combined through a genetic algorithm that adaptively searches for optimal ensemble weights. The approach was evaluated across five independent experimental executions to ensure its robustness and generalizability. Among these, the most balanced configuration achieved the lowest Root Mean Squared Error (RMSE) of 0.5460 and Mean Squared Error (MSE) of 0.2981, demonstrating a notable improvement in predictive accuracy. Comparative analysis further demonstrated that the genetic algorithm-optimized ensemble outperformed individual base learners and traditional ensemble methods by improving model synergy and reducing overfitting. These findings demonstrated how well evolutionary weight optimization works to improve ensemble predictions for evaluating plant diseases. The proposed framework provides a useful, adaptable approach that can be used with various foliar disease-prone crops and agroecological settings. This study advances intelligent crop monitoring systems, promoting sustainable agriculture and better disease management decision-making by facilitating earlier and more accurate disease severity estimation. The results highlight the possibility of combining genetic algorithms and ensemble learning in creating high-performing prediction tools that address important precision agriculture challenges.</p> Graphical Abstract <p>The graphical abstract illustrates the integrated approach for assessing leaf disease severity in Harumanis mango trees using ensemble machine learning models optimized through genetic algorithms. The study begins with the identification of foliar diseases in Harumanis mango plantations and the collection of visual data from leaves. These leaf images undergo segmentation and analysis to extract morphological features that serve as predictive variables. The modeling framework incorporates several regression algorithms, including Support Vector Machine, Gradient Boosting, and XGBoost, as base learners. These learners are then combined using a genetic algorithm that determines the optimal weights for ensemble predictions. The modelling flowchart highlights the evolutionary processes of selection, crossover, and mutation, which enable the genetic algorithm to fine-tune ensemble contributions. A series of performance plots showcases the predictive outcomes, indicating improved generalization and minimized error metrics, with the best model achieving an RMSE of 0.5460. The results validate the effectiveness of genetic algorithm-driven ensemble learning for disease prediction, while the adaptability of the framework makes it suitable for broader applications in precision agriculture and plant health monitoring.</p> <p></p>