Malicious Package Detection on PyPI
99.5AccuracyPyRadar
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| PyRadarAnalysis Type=Hybrid, Model=ML (RF), Feature Analysis=Statistical+feature importance, Detection Model Stability=×, Explainability (Per Package)=×2026.04 | 99.5 | — | — | |
| eDySecAnalysis Type=Dynamic, Model=DL (MLP), Feature Analysis=Statistical+ FLAML+ Probability, Detection Model Stability=✓, Explainability (Per Package)=✓2026.04 | 99 | — | — | |
| PypiGuardAnalysis Type=Hybrid, Model=Meta (Stacking), Feature Analysis=Statistical + ML-based, Detection Model Stability=✓, Explainability (Per Package)=×2026.04 | 98.43 | — | — | |
| PYGUARDAnalysis Type=Hybrid, Model=RAG-based LLM, Feature Analysis=Semantic + taxonomy, Detection Model Stability=×, Explainability (Per Package)=×2026.04 | 97.37 | — | — | |
| CLAMPD-NetAnalysis Type=Hybrid, Model=DL (CNN-BiGRU), Feature Analysis=Fusion + graph + semantic, Detection Model Stability=Partial, Explainability (Per Package)=Partial2026.04 | 97.23 | — | — | |
| Ladisa et al.Analysis Type=Static, Model=ML (XGB), Feature Analysis=Statistical+domain expertise, Detection Model Stability=Partial, Explainability (Per Package)=×2026.04 | 97 | — | — | |
| PyGuardEXAnalysis Type=Static, Model=ML (RF), Feature Analysis=Statistical+feature importance, Detection Model Stability=×, Explainability (Per Package)=×2026.04 | 97 | — | — | |
| DySecAnalysis Type=Dynamic, Model=ML (RF), Feature Analysis=Statistical + ML, Detection Model Stability=×, Explainability (Per Package)=×2026.04 | 95.99 | — | — | |
| CerebroAnalysis Type=Static, Model=DL (RoBERTa), Feature Analysis=Sequence modeling, Detection Model Stability=×, Explainability (Per Package)=×2026.04 | — | — | 96.1 | |
| Ea4mpAnalysis Type=Static, Model=DL + Ensemble, Feature Analysis=Code behavior sequences, Detection Model Stability=Partial, Explainability (Per Package)=×2026.04 | — | 97.6 | — | |
| MalGuardAnalysis Type=Static, Model=ML (RF), Feature Analysis=Graph + ML, Detection Model Stability=×, Explainability (Per Package)=Partial2026.04 | — | 99 | — | |
| OSCARAnalysis Type=Dynamic, Model=Rule-based, Feature Analysis=Behavioral feature, Detection Model Stability=×, Explainability (Per Package)=×2026.04 | — | 91 | — | |
| PyPIMalDetAnalysis Type=Hybrid, Model=ML (Stacking), Feature Analysis=DAE + stacking fusion, Detection Model Stability=×, Explainability (Per Package)=×2026.04 | — | 97.91 | — | |
| Samaana et al.Analysis Type=Static, Model=ML (Stacking), Feature Analysis=Statistical+feature importance, Detection Model Stability=Partial, Explainability (Per Package)=×2026.04 | — | 94.2 | — |