Table 1 The classification error metrics for Grace, SevenNet and CHGNet across different phases

From: Machine learning potentials for alloys: a detailed workflow to predict phase diagrams and benchmark accuracy

Model

Phase

Accuracy

Precision

Recall

F1

Grace

D1a

0.935

0.707

0.996

0.827

 

D019

0.945

0.796

1

0.886

 

FCC

0.864

0.892

0.445

0.616

 

HCP

0.877

0.897

0.819

0.857

 

Liquid

0.931

0.831

1

0.908

SevenNet

D1a

0.892

0.605

1

0.754

 

D019

0.705

0.415

1

0.591

 

FCC

0.823

1

0.313

0.478

 

HCP

0.840

0.947

0.682

0.793

 

Liquid

0.601

0.833

0.780

0.806

CHGNet

D1a

0.825

0

0

-

 

D019

0.622

0.280

0.480

0.355

 

FCC

0.778

0

0.091

0.167

 

HCP

0.746

0.794

0.585

0.674

 

Liquid

0.968

0.916

0.997

0.954

Back to article page