Machine Learning-Guided Design and Characterisation of Coconut Shell Pyrolyzed Particles-Reinforced High Entropy Alloy Composites

This study presents a novel approach to enhancing High-Entropy Alloy (HEA) composites reinforced with Coconut Shell Pyrolyzed (CSP) particles by integrating machine learning techniques and innovative material design. This work presents an initial assessment of phase formation prediction in High Entropy Alloys using three fundamental machine learning algorithms-Support Vector Machine (SVM), Random Forest (RF), and eXtreme Gradient Boost (XGBoost)-with experimental validation. Comparative analysis reveals that Valence Electron Concentration (VEC) and mixing enthalpy (ΔHmix) as the most influential parameters. The methodology involves the fabrication of AlCoCrFeNi-HEA and HEA-based composites reinforced with 1 wt% and 2 wt% CSP particles using powder metallurgy and Spark Plasma Sintering (SPS) techniques. The reliability of the machine learning model is validated through experimental characterization. XRD analysis reveals a dual-phase (FCC + BCC) structure in the sintered composites, which aligns consistently with the predicted phase structures from the RF model. Test accuracy showed comparable results for SVM (0.7838), RF (0.7838), and XGBoost (0.7770). Cross-validation demonstrated superior generalisation by XGBoost (0.8123 ± 0.0362) and RF (0.8038 ± 0.0369), while SVM showed greater variability (0.7633 ± 0.0580). RF achieved the highest precision (0.8767), followed by XGBoost (0.8616) and SVM (0.8417). Among the three classical ML algorithms tested in this preliminary study (SVM, Random Forest, and Decision Tree), Random Forest showed the highest prediction accuracy.