venkateshrr/protein-classification-enterprise-v2

Protein Classification Enterprise

Overview

Enterprise-grade protein classification system trained on 45 different deep learning architectures.

Best Model Performance

Model: Swin-Small

• Validation Accuracy: 1.0000
• Validation F1 Score: 1.0000
• Parameters: 1,223,946
• Best Epoch: 39

Ensemble Performance

Best Ensemble: Soft Voting

• Ensemble F1 Score: 1.0000
• Ensemble Accuracy: 1.0000

Top 5 Models

Rank	Model	F1 Score	Accuracy	Parameters
1	Swin-Small	1.0000	1.0000	1,223,946
2	ProteinAttention	1.0000	1.0000	299,274
3	Swin-Tiny	0.9987	0.9974	827,402
4	ResNet18	0.9973	0.9930	3,853,834
5	ResNet34	0.9947	0.9904	7,231,882

Training Details

• Total Models Trained: 45
• Total Training Time: 3.15 hours
• Framework: PyTorch
• Device: cuda
• Batch Size: 64
• Epochs: 50
• Optimizer: AdamW

Model Categories

• CNN Models (8): ResNet, VGG, DenseNet
• Transformer Models (6): ViT, Swin, BERT
• Efficient Models (8): EfficientNet, MobileNet, ShuffleNet
• Advanced Models (8): Inception, SENet, RegNet
• Protein-Specific Models (6): AAC, DPC, PseAAC, PCPE, CNN, Attention
• Custom Hybrid Models (9): CNN-RNN, CNN-Transformer, Multi-Scale
• Custom RNN Models (3): LSTM, GRU, Bidirectional

Results Files

• results/training_results.csv - Training metrics for all models
• results/test_results.csv - Test set performance
• results/ensemble_results.csv - Ensemble method comparison
• results/results_20251213_164532.json - Detailed results in JSON format