File size: 2,863 Bytes
71d3a93 6f766a1 71d3a93 6f766a1 71d3a93 6f766a1 71d3a93 6f766a1 71d3a93 6f766a1 71d3a93 6f766a1 71d3a93 6f766a1 71d3a93 6f766a1 71d3a93 6f766a1 71d3a93 6f766a1 71d3a93 6f766a1 71d3a93 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 | ---
license: other
library_name: scikit-learn
pipeline_tag: tabular-classification
tags:
- cybersecurity
- vulnerability-detection
- defect-prediction
- random-forest
- scikit-learn
- joblib
- nasa-promise
---
# Agent 1 — Structural Risk Random Forest
Random Forest that estimates defect risk from 21 NASA/PROMISE code metrics.
Part of a three-agent DevSecOps auditor (metrics → transformer → gatekeeper).
## Model details
| Field | Value |
|---|---|
| Version | 1.0.0 (initial release) |
| Artifact | `model.joblib` (bundle: `model`, `raw_cols`, `feature_cols`, `threshold`) |
| Task | Binary tabular classification (defect / no defect) |
| Threshold | **0.3183** (tuned on training CV with F2) |
| Training data | NASA/PROMISE PC1 (`pc1.csv`) |
| Dataset version | Not pinned — static PC1 snapshot, no revision hash recorded |
| Seed | 42 |
`metadata.json` mirrors the dataset, threshold, and best params.
## Input features
21 numeric metrics:
```
loc, vg, evg, ivg, n, v, l, d, i, e, b, t,
locode, locomment, loccodeandcomment, loblank,
uniq_op, uniq_opnd, total_op, total_opnd, branchcount
```
Five derived features are added before prediction. Final `feature_cols` order
(26 columns, must match exactly):
```
loc, vg, evg, ivg, n, v, l, d, i, e, b, t,
locode, locomment, loccodeandcomment, loblank,
uniq_op, uniq_opnd, total_op, total_opnd, branchcount,
complexity_density, effort_per_loc, comment_ratio, branch_density, operand_reuse
```
## Preprocessing
- All 21 inputs must be numeric; no missing values (nothing is imputed).
- Derive ratios with `+1` denominators:
```
complexity_density = vg / (loc + 1)
effort_per_loc = e / (loc + 1)
comment_ratio = locomment / (loc + 1)
branch_density = branchcount / (loc + 1)
operand_reuse = total_opnd / (uniq_opnd + 1)
```
- Select `b["feature_cols"]` in the order above.
## Output
Raw: `predict_proba(X)[0, 1]` — defect risk in `[0, 1]`, uncalibrated.
`high_risk = score >= 0.3183`.
Reference wrapper output:
```json
{"struct_risk_score": 0.5105, "high_risk": true, "threshold": 0.3183, "status": "ok", "reason": ""}
```
## Usage
```python
import joblib
import pandas as pd
b = joblib.load("model.joblib")
df = pd.DataFrame([metrics])[b["raw_cols"]].apply(pd.to_numeric)
df["complexity_density"] = df["vg"] / (df["loc"] + 1)
df["effort_per_loc"] = df["e"] / (df["loc"] + 1)
df["comment_ratio"] = df["locomment"] / (df["loc"] + 1)
df["branch_density"] = df["branchcount"] / (df["loc"] + 1)
df["operand_reuse"] = df["total_opnd"] / (df["uniq_opnd"] + 1)
X = df[b["feature_cols"]]
score = b["model"].predict_proba(X)[0, 1]
high_risk = score >= b["threshold"]
```
Trained on NASA PROMISE PC1. Metrics from other languages or domains are
out-of-distribution.
## License
Not specified. Update the `license` field before publishing.
|