Implementation of Tree based Machine Learning Model to Predict Stability of Multicomponent Materials of B and Si Using Compositional-based Features

The stability of materials plays a crucial role in expediting the development in material science and engineering. The stability factor determines the synthesizability of the material. The utilization of Density Functional Theory (DFT) enables the examination of material stability; however, the process is hindered by the high computational costs and time-consuming calculations, making it challenging to forecast the stability of numerous potential materials. In this work, a machine learning model (MLM) was developed to anticipate material stability by taking into account compositional-based features. A total of 8763 multicomponent materials of B and Si from the material project database were used in this study. The Elemental Property compositional-based featurizer from the MATMINER packages was employed to create 133 new features, out of which only 25 were chosen using the forward selection technique. The dataset was divided into training and testing sets with a test size of 0.2 for model training. The model was trained, and hyperparameters were fine-tuned through 10-fold cross-validation using the Scikit library in the Anaconda distribution. The Random Forest Classifier and Gradient Boosting Classifier exhibit noteworthy accuracies of 0.873 with F1 scores of 0.851 and 0.867, respectively, on the test data. Conversely, the Extra Trees Classifier demonstrates slightly inferior performance compared to the aforementioned models, yet it achieves a satisfactory F1 score of 0.80 and an accuracy of 0.849. Our investigation demonstrates the potential of machine learning models in predicting material stability thus aiding researchers in expediting material discoveries.