Implementación de un algoritmo de clasificación para la asignación del sector destino en importaciones del mercado ecuatoriano.

Abstract

This research aimed to develop and evaluate a classification model based on machine learning techniques for assigning the destination sector of imports in the Ecuadorian market. The study follows a quantitative categorical approach, focusing on classifying records into specific categories (Aquaculture, Livestock, Chemical, and Veterinary) through the analysis of historical data obtained from the COBUS platform and internal company records. Methodologically, a literature review was conducted on the main machine learning approaches applicable to classification problems, emphasizing supervised models and, particularly, tree-based methods. Subsequently, data preparation and cleaning processes were carried out to ensure quality, coherence, and consistency through data filtering, removal of irrelevant variables, and standardization procedures. The main implemented model was Random Forest, complemented by a Decision Tree model for comparative purposes. The dataset was divided into 75% for training and 25% for testing, using stratification to maintain class proportions. Model performance was evaluated through accuracy metrics, classification reports, confusion matrices, and multiclass ROC curves to measure discriminative capacity. The results demonstrated that the Random Forest model achieved greater stability and predictive accuracy compared to the individual Decision Tree model. Additionally, feature importance analysis identified the most influential variables in the classification process, while the detection of low-probability records contributed to data quality assessment.