Malicious Package Detection on PyPI Small Packages N=972 (test)
97.31AccuracyH2GLM
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| H2GLM2026.07 | 97.31 | 95.68 | 99.08 | 97.35 | |
| Qwen3-14b2026.07 | 94.84 | 98.03 | 91.51 | 94.66 | |
| HERCULE2026.07 | 92.54 | 93.3 | 91.67 | 92.48 | |
| MPHunter2026.07 | 91.1 | 86.2 | 93.5 | 89.7 | |
| DeepSeek-R1-70b2026.07 | 83.43 | 76.19 | 97.25 | 85.44 | |
| Warehouse2026.07 | 79.78 | 94.61 | 63.17 | 75.76 | |
| OSSGadget2026.07 | 68.21 | 58.15 | 94.24 | 71.92 | |
| Naive Bayes2026.07 | 55.14 | 54.59 | 61.11 | 57.66 | |
| Qwen2.5-7b2026.07 | 48.17 | 47.93 | 42.55 | 45.08 | |
| Bandit4Mal2026.07 | 48.15 | 42.1 | 26.5 | 32.53 | |
| Llama3-8b2026.07 | 45.3 | 46.79 | 68.58 | 55.63 | |
| Clustering2026.07 | 36.96 | 23.11 | 11.21 | 15.09 |