Advanced machine learning assisted prediction of multi-fiber hybrid natural fiber/epoxy composites based on jute, coconut coir, banana, and pineapple leaf fibers

A growing demand for lightweight, high-performance and ecofriendly engineering materials, the use of renewable lignocellulosic materials as environmentally friendly substitutes to synthetic reinforcements has been paid considerable attention. In the present work, mechanical, thermal, morphological and machine learning based prediction properties of hybrid natural fiber reinforced epoxy composites are investigated which are prepared using alkaline treated jute fiber (JF), coconut coir (CC), banana fiber (BaF) and pineapple leaf fiber (PALF). Six hybrid composite materials containing a mixture of these 4 natural fibers were made using a hand lay-up technique with a constant 50 wt. % fiber loading. Tensile, flexural, impact and hardness tests were conducted on the composites, along with scanning electron microscopy (SEM), Fourier transform infrared spectroscopy (FTIR), atomic force microscopy (AFM), dynamic mechanical analysis (DMA), and thermogravimetric analysis (TGA) to study their morphological, chemical, viscoelastic, surface and thermal properties. The surface roughness of fiber was also found to be significantly improved by alkaline treatment (5 wt. % NaOH), which increased the AFM average roughness (Ra) from 12.60 nm to 31.43 nm, which led to a better fiber–matrix interfacial adhesion. Sample S-5 (20 wt.% kJF + 20 wt.% CC + 10 wt.% PALF) showed the best overall performance with tensile strength of 85.8 MPa, flexural strength of 134.5 MPa, impact strength of 23.3 kJ/m 2 , and the Shore D hardness of 72.6. SEM images showed better bonding between the fibers and matrix and less pull-out, DMA showed higher storage modulus and a glass transition temperature of 65°C, TGA showed higher thermal stability with onset degradation temperature of 225°C and residual char yield of 13.7%. In order to investigate the predictive modeling ability, four regression algorithms were considered: Linear Regression (LR), Multiple Linear Regression (MLR), Decision Tree Regression (DTR), and Random Forest Regression (RFR), which were tested with the experimentally obtained dataset. Of these models, the RFR model gave the best predictive performance with R 2 values of 0.968, 0.939, 0.941 and 0.962 for tensile strength, flexural strength, impact strength and hardness, respectively. The limited number of data makes the results of the machine learning only a proof of concept for the possibility of data-driven prediction for hybrid natural fiber composites. The results demonstrate how strategic fiber hybridization can enhance the properties of composites and offer a basis for future research using more data and more sophisticated machine learning models to improve the design of sustainable composite materials.