A Genetic Algorithm-Optimized Kernel Density Estimation and D-S Evidence Fusion Classification for Predicting the...

Area of Science:

• Materials Science
• Data Science
• Artificial Intelligence

Background:

• Accurate material stability classification often requires fusing multiple features under uncertainty.
• Dempster-Shafer (D-S) theory is a powerful framework for multi-source information fusion but relies on heuristic basic probability assignments (BPAs).

Purpose of the Study:

• To develop a data-driven classification model for material stability prediction that overcomes limitations of heuristic BPA assignment in D-S theory.
• To integrate genetic algorithm (GA)-optimized kernel density estimation (KDE) with a weighted D-S fusion strategy for improved BPA construction.

Main Methods:

• A novel classification model integrating GA-optimized KDE with a weighted D-S fusion strategy was proposed.
• GA was used to automatically select optimal kernel functions and bandwidth for KDE, enabling data-driven BPA construction.
• The model was validated on benchmark datasets and applied to predict the thermodynamic stability of double perovskite halide materials.

Main Results:

• The GA-KDE-DS framework achieved competitive or superior performance on benchmark datasets.
• For double perovskite halide materials, the model attained 93.7% accuracy and 85.3% precision, outperforming existing methods.
• The model demonstrated robust performance and strong transferability, even with chemical elements absent from the training set.

Conclusions:

• The proposed GA-KDE-DS framework offers a practical solution for material stability classification under uncertainty and limited data.
• This approach facilitates accurate BPA construction without manual parameter tuning, enhancing the reliability of D-S theory.
• The method shows significant potential for accelerating the discovery of novel functional materials.