Prediction of Hygroscopic Properties of Untreated Borassus Fiber Using Ensemble Learning Techniques

Abstract

—Borassus palm or aethiopium palmyra is a palm species tree, widely spread in sub-Saharan Africa but its fruits don’t have any economic value therefore considered as waste. This study investigated the potential of Borassus fruit fibers (BFF), extracted manually from the underutilized fruit, for various applications by examining their hygroscopic properties. Scanning electron microscopy (SEM) revealed the fibers' unique features, including a relatively large diameter and high affinity for water vapor. A Dynamic Vapor Sorption (DVS) analysis with exposure time varying from 1, 2, 4 until 72h and varied Relative Humidity (from 0 to 90%) with 10% increment was carried out to examine the Sorption-desorption behavior. The characteristic hysteresis behavior of natural fibers was observed, with significant moisture uptake, particularly above 70% RH. The sorption and desorption processes were quantified, revealing a linear relationship between mass change and relative humidity. Furthermore, an Ensemble learning approach, specifically a Gradient Boosting Regression (GBR) model, was developed to predict the hygroscopic behavior of BFF. Trained on the experimental sorption-desorption data, the GBR model demonstrated excellent predictive accuracy, achieving a high R² value of 91.7% and low CV, MSE, and RMSE values (6.9 and 2.6, respectively). These findings highlight the significant influence of relative humidity on BFF moisture content and demonstrate the effectiveness of GBR as a powerful tool for accurately predicting the complex hygroscopic behavior of these fibers. Keywords— Machine Learning techniques, Ensemble Learning, Natural fiber, Hygroscopic properties, sorption/desorption