Table 1 Comparison of average accuracy among different models under 10-fold cross-validation.
From: A general and transferable deep learning framework for predicting phase formation in materials
Model | Data representation | Input size | Algorithms | Average accuracy | |
|---|---|---|---|---|---|
Training | Testing | ||||
Ward’s work7 | Manual features vector generated by Magpie | 145 | Random forest | – | 90% |
SNN1 | Manual features vector + processing parameter | 13 + 1 | SNN | 89.8% | 89.9% |
SNN2 | Composition vector + processing parameter | 73 + 1 | SNN | 93.2% | 92.8% |
SNN3 | Manual features vector + composition vector + processing parameter | 86 + 1 | SNN | 93.9% | 93.5% |
SNN4 | Manual features vector generated by Magpie + processing parameter | 145 + 1 | SNN | 90.1% | 90.0% |
CNN1 | Atom table representation | 11 × 11 | CNN | 96.4% | 95.0% |
CNN2 | Randomized periodic table representation | 9 × 18 | CNN | 96.7% | 94.9% |
CNN3 | Periodic table representation | 9 × 18 | CNN | 96.4% | 96.3% |
