Predicting Wood Color Origin Using Random Forest and Spectroscopy: A Machine Learning Approach for Industrial and Forestry Applications

Quinteros-Navarro D.; Quinteros A.; Muñoz V.; Ramirez W.; Aguilar J.; Pezo M. · Ingenierie des Systemes d'Information (2025) · DOI: 10.18280/isi.300905

Resumen

Wood is a natural resource highly valued in industrial, commercial, and decorative sectors, where color is a critical property for classification and market acceptance. Color reflects both the anatomical structure and chemical composition of wood, being strongly influenced by phenolic compounds and environmental conditions. This study develops a predictive model to determine the origin of wood color using a Random Forest regressor applied to a dataset of 300 samples. The dataset integrates physicochemical (lignin, flavonoids, tannins, anthocyanins, polyphenols, pH) and environmental variables (temperature, humidity, precipitation, altitude). Preprocessing included categorical encoding, normalization, and generation of derived variables such as pigmentation level and pigmentation time. The Random Forest model achieved high predictive accuracy with MAE = 0.14, RMSE = 0.22, and R2= 0.89. The most influential variables were flavonoids, tree age, temperature, and internal humidity. Results confirm that higher temperatures accelerate pigmentation timing, while species with slower growth rates show later pigmentation. The proposed machine learning approach demonstrates the feasibility of objectively predicting wood color origin based on physicochemical and environmental data. This contributes to industrial wood classification, quality control, and resource valorization by reducing subjectivity, lowering costs, and improving consumer satisfaction.

Instituciones: Universidad Tecnológica del Perú; Universidad Nacional de San Martín