Prediction Model for Cacao Production Integrated into an Offline Mobile Application: The Impact of Artificial Intelligence on Agricultural DecisionMaking

Abstract

Cacao production, a key economic pillar for numerous rural communities in Ecuador, faces structural challenges related to climate variability and limited digital connectivity. This study presents the development and implementation of a yield prediction model based on the XGBoost algorithm, integrated into an offline mobile application designed to operate in agricultural environments without internet access. The research followed the CRISP-DM methodology and included the analysis of 5584 observations collected from plots in La Maná (Cotopaxi), corresponding to three cacao genotypes. Variables were processed using cleaning, imputation, and normalization techniques. The predictive model, validated with standard metrics (MSE, RMSE) and an $R^{2}$ of 0.9399, demonstrated robust fit and high interpretability. Subsequently, the model was deployed in a mobile app developed with React Native. Field deployment showed response times under five seconds, compatibility with low-end devices, and high user acceptance. Participatory validation confirmed the practical usefulness of the tool for real-time agronomic decision-making. This work provides evidence of the value of AI tailored to rural contexts and proposes a replicable approach for other value chains under similar conditions.