Optimizable Machine Learning Neural Network Technique Used for Predicting Transition Metal Properties: Debye Temperature and Thermal Conductivity

With the advent of the big data era, artificial intelligence technology has penetrated and deeply affected our daily lives. In addition, optimizable machine learning neural network techniques enable computers to learn from transition metal data without explicit programming. This study aims to predict the Debye temperature and lattice thermal conductivity of transition metals from the dataset (981 records) using their physical and chemical properties, with the goal of designing advanced materials. Traditional methods for estimating these properties were often time-consuming and computationally intensive, prompting the emergence of neural network-based machine learning techniques as a promising alternative. Among the trained models, the Wide Neural Networks (WNN) demonstrated strong predictive performance (R² values) of 0.985 and 0.977 on the validation sets and 0.994 and 0.989 on the test sets for the Debye temperature and thermal conductivity, respectively. A significant correlation between the Debye temperature and thermal conductivity was identified, and partial dependence plots were used to interpret feature effects and validate model predictions. These findings demonstrate the effectiveness of neural network-based machine learning techniques in predicting thermophysical properties of transition metals.