Experimental Dataset and Predictive Model for Multi-Product Yield Prediction from Cocoa Residues

This repository contains the complete experimental dataset (84 laboratory trials), the trained Artificial Neural Network model, and an interactive Google Colab notebook developed for the valorization of cocoa residues (Theobroma cacao L.) in Santander and Norte de Santander, Colombia. The resources support the research article "A hybrid empirical-AI model for multi-product yield prediction and stochastic optimization from cocoa residues" published in Sustainability (2026). The dataset includes operational variables (temperature, time, pH, solvent concentration, extraction method, cocoa variety) and yields for five target compounds: bioethanol, essential oils, antioxidants, pectins, and paraffins. The predictive model integrates a hybrid empirical-AI architecture with dimensional consistency, trained on 100% empirical data, combined with Monte Carlo sensitivity analysis (10,000 simulations) for uncertainty quantification.