PREDICTION OF QUANTUM DESCRIPTOR VALUES IN BORON NITRIDE NANOCAGE-BASED MATERIALS USING ANN, CNN, AND RNN ALGORITHMS

• Main Author: Pradila, Rike
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/86934
• Institution: Institut Teknologi Bandung

This research aims to develop a method for predicting quantum descriptor values, specifically the HOMO-LUMO gap, in larger boron nitride nanocage materials using a machine learning approach. The dataset was obtained through DFT calculations on B12N12, B24N24, and B36N36 nanocages to determine the HOMOLUMO gap values as prediction targets. Three ML models ANN, CNN, and RNN were employed to evaluate their ability to learn patterns from smaller nanocages (B12N12 and B24N24) and generalize to predict the HOMO-LUMO gap for a larger nanocage (B36N36) excluded from the training dataset. The results show that RNN achieved the best performance with an RMSE of 0,9289, MSE of 0,8629, MAE of 0,6457, and R² of 0,91. CNN demonstrated significant capability in capturing nonlinear relationships between structural features and electronic properties, achieving an RMSE of 0,8115, MSE of 0,6585, MAE of 0,7292, and R² of 0,73. In contrast, ANN showed the lowest performance, with an RMSE of 0,4746, MSE of 0,225, MAE of 0,3805, and R² of 0,65, reflecting its limitations in capturing complex patterns in nanocage datasets with limited features. These findings are expected to serve as a foundation for developing more efficient and accurate ML-based prediction methods for designing boron nitride nanocages of greater size and complexity. By leveraging ML approaches such as RNN and CNN, this study highlights the potential to reduce the computational time and cost associated with theory-based methods, particularly in applications requiring rapid analysis and high scalability for functional material exploration.