Machine learning guided prediction of electronic and magnetic properties of 2D transition metal dichalcogenide alloys

• Main Author: Jain, Aarushi
• Other Authors: Liu Zheng
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2023
• Subjects: Engineering::Materials
• Online Access: https://hdl.handle.net/10356/166730
• Institution: Nanyang Technological University

In this work, machine learning is used to predict the electronic, magnetic and thermodynamic properties of 2-dimensional transition metal dichalcogenides (TMD) and its alloys. TMDs have attracted great interest due to their highly applicability in modern-day transistors, sensors, photodetectors and more. The heat of formation, energy above convex hull, bandgap, direct bandgap and magnetic moment are the properties that were predicted, with the highest R squared value of 0.97. Over 60 features and 1147 2D materials with their target properties were extracted from Python’s Mendeleev package and C2DB respectively. The models used for training were Multilayer Perceptron, Extreme Gradient Boosting and Support Vector Machine; model selection was done using cross validation with hyperparameter tuning. Feature engineering, importance and dimensionality reduction were found to improve the model performance significantly. Finally, the best models and training sets were used for prediction on new test data from 2Dmatpedia database, demonstrating good performance. The work shows great potential to be extended in the future for discovery of novel materials from search space, which could be synthesised in labs and confirmed.