Banana Quality Classification Using Explainable Machine Learning: A Perspective for the Ecuadorian Export Sector

Abstract

Context: This study analyzes the most relevant characteristics associated with banana quality using machine learning techniques, within the context of Ecuador as the world's leading exporter of this fruit, where local scientific production on this subject remains limited. Method: The CRISP-DM methodology was applied to the public “Banana Quality Dataset,” using three supervised classification algorithms: XGBoost, Naive Bayes, and CN2 Rule Induction, evaluated using stratified 10-fold cross-validation. Results: The results indicate that the maturity index, sugar content, and fruit length constitute the most decisive predictors of quality. In terms of evaluation, CN2 Rule Induction and XGBoost achieved accuracy and F1-score values of 85%, while Naive Bayes presented a competitive performance with values exceeding 80%. The use of explainable artificial intelligence (XAI) tools allowed for the extraction of interpretable rules that provide transparency to the classification process. Conclusions: This study contributes to narrowing the existing research gap in producing countries like Ecuador and demonstrates the viability of implementing data science solutions using open-source tools.