Regresar

Hexagonal Architecture and Machine Learning for Cocoa Disease Management in a Mobile App

Abstract:

This study presents a technological solution for the early detection of moniliasis in cocoa plantations in rural Ecuador, through a hybrid mobile application based on hexagonal architecture and machine learning. Moniliasis, caused by Moniliophthora perniciosa, can lead to crop losses of up to 9 0% in key regions, posing a serious threat to food security and rural livelihoods. The proposed approach combines the PRISMA methodology for a systematic literature review and the CRISP-DM framework for developing predictive models, using data from environmental sensors and manual field records. Dimensionality reduction techniques such as PCA and linear KPCA showed the highest accuracy (0. 8 3 7 6). For regularization, Linear and Ridge regressions outperformed Lasso and ElasticNet. Ensemble learning methods, particularly Random Forest (0.9888) and Gradient Boosting (0.9805), achieved the best classification results. Model validation using K-Folds and Randomized Search confirmed the system's robustness. Additionally, REST API load testing with 500 simulated POST requests demonstrated stable performance and reliable response times, making the system viable for deployment in rural-like conditions. Nonetheless, some limitations were identified, including the absence of field validation, limited agronomic metadata, and lack of direct user interaction. Future work should incorporate usability testing with farmers, multilingual support, and partnerships with local institutions to enhance adoption and contextual relevance. In conclusion, this research offers a solid foundation for smart agriculture applications in Latin America. By refining the system through participatory approaches and contextual integration, the proposed solution has the potential to transform disease management in smallholder farming systems through scalable, accurate, and locally-adapted digital tools.