Ensemble Learning for Accurate Mineral Structure Prediction using CONVNET and Material Data

Predicting mineral structures at a given chemical composition before conducting any experimental practices is crucial. This paper proposes an advanced ensemble learning algorithm for accurate prediction of mineral structure using a Convolutional Neural Network (ConvNet) and material data. This approach follows a data processing approach in which raw material data undergoes a data cleaning and data wrangling process, and noise or impurities present in there are removed. The data is graphically represented by using a heat map. The feature transformation and feature scaling techniques are employed to normalize the features of the material data. This enhances the convergence of the model during training. The ConvNet is integrated to automatically predict the structure of the material from the raw data. The hyperparameters of the ConvNet are finely tuned by the Crow Search Optimization (CSO) algorithm. The CSO enhances both the accuracy and generalization of the proposed model. The integration of ConvNet with CSO technique exhibits exceptional performance in predicting the mineral structure accurately compared with existing techniques. The proposed approach is implemented in Python software and validates the accuracy of $82 \%$ thus, ensuring a promising tool for material design and discovery.