Predictive modelling of compressive strength in silicon nitride-reinforced aluminium composites using supervised machine learning: A comparative study of random forest and artificial neural networks

The integration of machine learning and material science is now changing the way property predictions are made, accelerating the process of discovery and optimization of advanced materials. For this research, we employed a supervised ML to estimate the compressive strength of AMCs reinforced by silicon nitride. To sharpen predictive accuracy, hyperparameter tuning using GridSearch CV was performed. We employed two different algorithms-RF and ANN-to unravel the complex links between the inputs, such as compaction pressure, reinforcement content, sintering temperature, and sintering time with the target strength. Regularization was used to guard against overfitting, and training versus testing performance was compared rigorously. The results indicated that the RF outperformed the ANN model, giving an R² of 0.88, while the ANN reached an R² of only 0.80. These results suggested that the RF model had tighter residuals and higher accuracy since it emphasizes the parameters that drive most of the variance in the dataset. This work illustrates the potential of ML, especially RF, in reliably predicting properties of materials and informing the design of high-performance composites.