Fiber Quality Prediction Using Nir Spectral Data: Tree-Based Ensemble Learning VS Deep Neural Networks

Abstract

The growing applications of near infrared (NIR) spectroscopy in wood quality control and monitoring necessitates focusing on data-driven methods to develop predictive models. Despite the advancements in analyzing NIR spectral data, literature on wood science and engineering has mainly utilized the classic model development methods, such as principal component analysis (PCA) regression or partial least squares (PLS) regression, with relatively limited studies conducted on evaluating machine learning (ML) models, and specifically, artificial neural networks (ANNs). This could potentially limit the performance of predictive models, specifically for some wood properties, such as tracheid width that are both time-consuming to measure and challenging to predict using spectral data. This study aims to enhance the prediction accuracy for tracheid width using deep neural networks and tree-based ensemble learning algorithms on a dataset consisting of 2018 samples and 692 features (NIR spectra wavelengths). Accordingly, NIR spectra were fed into multilayer perceptron (MLP), 1 dimensional-convolutional neural networks (1D-CNNs), random forest, TreeNet gradient-boosting, extreme gradient-boosting (XGBoost), and light gradient-boosting machine (LGBM). It was of interest to study the performance of the models with and without applying PCA to assess how effective they would perform when analyzing NIR spectra without employing dimensionality reduction on data. It was shown that gradient-boosting machines outperformed the ANNs regardless of the number of features (data dimension). All the models performed better without PCA. It is concluded that tree-based gradient-boosting machines could be effectively used for wood characterization utilizing a medium-sized NIR spectral dataset.

Utilization of Synthetic Near-Infrared Spectra via Generative Adversarial Network to Improve Wood Stiffness Prediction

“…Training on tabular data, the general superior performance of models based on the gradient-boosted decision tree over deep learning methods has been reported, specifically where machine learning models outperformed deep learning models in regression [59]. Nasir et al (2023) [27] showed that tree-based gradient-boosting machines such as LGBMs, XGBoost, and TreeNet outperformed the ANN and CNN models when predicting fiber properties using NIR spectral data (with and without applying PCA). Thus, one might speculate that the LGBM model was so robust on the original training dataset that it did not experience significant improvement in its performance by changing the size of the training data.…”

“…ANNs are biologically inspired mathematical models that can explain variations in almost any type of dataset with a good degree of accuracy. Therefore, these are one of the most widely used deep learning neural networks for regression and classification [ 27 ]. ANNs consist of input, hidden, and output layers, with the layers consisting of neurons that are interconnected by weighted links [ 50 ].…”

“…The LGBM technique employs a histogram-based decision-tree-learning technique that optimizes memory utilization and reduces communication overhead. Numerous applied machine learning tasks have used LGBM techniques because of its excellent predictive power, effectiveness, and capacity for handling complex datasets [ 27 , 54 ]. The algorithm has several key parameters that control overfitting, complexity, and the optimization process.…”

“…Research on using machine learning and deep learning applied to NIR spectra for wood characterization and monitoring is relatively limited. Studies on small-sized NIR spectral datasets (sample sizes ranging from 172 to 480) showed artificial neural networks (ANNs) outperformed PLS regression models [ 24 , 25 , 26 , 27 ]. Specifically, Ayanleye et al (2021) [ 26 ] used 240 samples to train ANN and neuro-fuzzy models to predict the MOE and MOR of western hemlock ( Tsuga heterophylla ) and Douglas-fir ( Pseudotsuga menziesii ) lumber.…”

Quality Control of Thermally Modified Western Hemlock Wood Using Near-Infrared Spectroscopy and Explainable Machine Learning

“…The spectra were classified using a TreeNet gradient boosting machine, a tree-based ensemble learning model. The model could successfully handle NIR spectra without the need for prior dimensionality reduction and was shown to outperform the ANN and convolutional neural network (CNN) for fiber quality prediction using NIR data [43]. Gradient boosting machines have been used in the wood science and technology literature for wood species identification [44] and predicting the properties of wood composites [45].…”