File size: 12,048 Bytes
c6a80e7 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 | """
feature_engineering.py
======================
Feature pipeline for the CYB003 baseline classifier.
Predicts `execution_phase` (10-class) from per-timestep malware execution
telemetry on the CYB003 sample dataset.
CSV inputs:
malware_samples.csv (primary, one row per timestep, 60 timesteps
per sample, 100 samples = 6000 rows)
sample_summary.csv (per-sample aggregates; reserved for future
work — joining inflates per-sample features
across 60 identical replications, which hurt
the model in pilot experiments)
environment_profiles.csv (reserved for future work)
execution_events.csv (reserved for future work)
Target classes (10 execution phases observed in the sample):
initial_drop, persistence_establishment, privilege_escalation,
lateral_movement, payload_execution, data_exfiltration,
c2_communication, dormancy_dwell, sandbox_evasion_stall,
self_destruct_cleanup
This corresponds to the SOC / sandbox-analyst use case: given the malware's
current behavioural state, what phase of execution is it in? Useful for
dynamic-analysis tools, EDR phase tagging, and behavioural classifiers.
The pivot to execution_phase (away from malware_family) happened because
malware family classification on n=100 samples with group-aware splitting
landed at majority-baseline accuracy (~15%, ROC-AUC ~0.58). execution_phase
sits on 6,000 rows of per-timestep data with strong, stable signal across
seeds (~91% accuracy, ROC-AUC ~0.98). See the model card for details.
Leakage analysis
----------------
No categorical feature has phase->phase purity above 0.17 (uniform random
baseline is 0.10), so nothing in the data is an oracle for the target.
The model relies on a mix of `timestep` (strong but not deterministic —
most phases have tight timestep windows, but `dormancy_dwell`,
`sandbox_evasion_stall`, and `self_destruct_cleanup` span the full
0-59 range) and behavioural features.
Public API
----------
build_features(samples_path) -> (X, y, groups, meta)
transform_single(record, meta) -> np.ndarray
save_meta(meta, path) / load_meta(path)
License
-------
Ships with the public model on Hugging Face under CC-BY-NC-4.0, matching
the dataset license. See README.md.
"""
from __future__ import annotations
import json
from pathlib import Path
from typing import Any
import numpy as np
import pandas as pd
# ---------------------------------------------------------------------------
# Label space
# ---------------------------------------------------------------------------
# Alphabetical for stable indexing.
LABEL_ORDER = [
"c2_communication",
"data_exfiltration",
"dormancy_dwell",
"initial_drop",
"lateral_movement",
"payload_execution",
"persistence_establishment",
"privilege_escalation",
"sandbox_evasion_stall",
"self_destruct_cleanup",
]
LABEL_TO_INT = {lbl: i for i, lbl in enumerate(LABEL_ORDER)}
INT_TO_LABEL = {i: lbl for lbl, i in LABEL_TO_INT.items()}
# ---------------------------------------------------------------------------
# Identifier and target columns - not features
# ---------------------------------------------------------------------------
ID_COLUMNS = ["sample_id", "family_id", "threat_actor_id"]
TARGET_COLUMN = "execution_phase"
# Note: malware_family is kept as a FEATURE for phase prediction (family
# is a useful observable - a SOC analyst knows what family they're looking
# at). It's not a leakage source for phase since phase->family purity is
# only 0.16. Same logic for threat_actor_tier, ep_stack, target_platform -
# these are environmental context, not oracles for phase.
# ---------------------------------------------------------------------------
# Per-timestep numeric features
# ---------------------------------------------------------------------------
DIRECT_NUMERIC_TIMESTEP_FEATURES = [
"timestep", # strong but non-deterministic phase signal
"api_call_rate",
"registry_write_count",
"network_connection_count",
"process_injection_flag",
"c2_beacon_interval_sec",
"av_signature_hit_flag",
"sandbox_evasion_flag",
"lateral_propagation_count",
"privilege_escalation_flag",
# PE static features (constant per sample but informative for phase
# given that the model sees these alongside per-step behaviour)
"pe_entropy_mean",
"pe_entropy_std",
"import_hash_cluster",
"section_count",
"packed_section_ratio",
"string_entropy_mean",
"byte_histogram_chi2",
"code_section_rx_ratio",
"resource_section_entropy",
"suspicious_import_count",
"packer_detected_flag",
]
CATEGORICAL_TIMESTEP_FEATURES = [
"malware_family", # kept as feature: phase prediction conditions
# on family (a known observable in SOC workflows)
"threat_actor_tier",
"target_platform",
"obfuscation_technique",
"detection_outcome",
"ep_stack",
]
# ---------------------------------------------------------------------------
# Engineered features (none derived from phase or timestep alone)
# ---------------------------------------------------------------------------
def _add_engineered_features(df: pd.DataFrame) -> pd.DataFrame:
"""
Six engineered features. None directly encode phase (that would be
a tautology); each is a behavioural composite that disambiguates
phases sharing similar timestep ranges.
"""
df = df.copy()
# 1. API burst score: high for execution-heavy phases (payload_execution,
# privilege_escalation), low for stealth phases (dormancy, evasion).
df["api_burst_score"] = (
df["api_call_rate"] * df["registry_write_count"].clip(upper=50)
).astype(float)
# 2. C2 active flag: positive c2_beacon_interval_sec indicates active
# beaconing. Strongly correlates with c2_communication phase.
df["is_c2_active"] = (df["c2_beacon_interval_sec"] > 0).astype(int)
# 3. High network volume step: above-threshold connection count, common
# in lateral_movement, data_exfiltration, c2_communication.
df["is_high_net_volume"] = (df["network_connection_count"] > 5).astype(int)
# 4. Stealth indicator: low api_call_rate AND no AV/sandbox hit. Used
# to disambiguate dormancy_dwell / sandbox_evasion_stall from active
# phases that happen to land in similar timestep windows.
df["is_stealth_step"] = (
(df["api_call_rate"] < 5)
& (df["av_signature_hit_flag"] == 0)
& (df["sandbox_evasion_flag"] == 0)
).astype(int)
# 5. Destructive action indicator: combines privilege escalation flag
# and registry-write count. High in persistence_establishment and
# self_destruct_cleanup.
df["is_destructive_step"] = (
(df["privilege_escalation_flag"] == 1)
| (df["registry_write_count"] > 10)
).astype(int)
# 6. Lateral activity: network connections combined with lateral_propagation
# count > 0. Distinguishes lateral_movement from other network phases.
df["lateral_activity_score"] = (
df["lateral_propagation_count"] * df["network_connection_count"]
).astype(float)
return df
# ---------------------------------------------------------------------------
# Public API
# ---------------------------------------------------------------------------
def build_features(
samples_path: str | Path,
) -> tuple[pd.DataFrame, pd.Series, pd.Series, dict[str, Any]]:
"""
Load CSV, drop identifier columns and target, engineer features,
one-hot encode, return (X, y, groups, meta).
`groups` is a Series of sample_id values aligned with X. Use it
with GroupShuffleSplit / GroupKFold: a single sample contains 60
correlated timesteps, and row-level random splitting inflates metrics.
"""
samples = pd.read_csv(samples_path)
# Extract target + groups
y = samples[TARGET_COLUMN].map(LABEL_TO_INT)
if y.isna().any():
bad = samples.loc[y.isna(), TARGET_COLUMN].unique()
raise ValueError(f"Unknown execution_phase values: {bad}")
y = y.astype(int)
groups = samples["sample_id"].copy()
# Drop target + identifiers from feature pool
samples = samples.drop(columns=ID_COLUMNS + [TARGET_COLUMN], errors="ignore")
# Engineered features
samples = _add_engineered_features(samples)
# Numeric features
numeric_features = (
DIRECT_NUMERIC_TIMESTEP_FEATURES
+ [
"api_burst_score", "is_c2_active", "is_high_net_volume",
"is_stealth_step", "is_destructive_step", "lateral_activity_score",
]
)
X_numeric = samples[numeric_features].astype(float)
# One-hot categoricals
categorical_levels: dict[str, list[str]] = {}
blocks: list[pd.DataFrame] = []
for col in CATEGORICAL_TIMESTEP_FEATURES:
if col not in samples.columns:
continue
levels = sorted(samples[col].dropna().unique().tolist())
categorical_levels[col] = levels
block = pd.get_dummies(
samples[col].astype("category").cat.set_categories(levels),
prefix=col, dummy_na=False,
).astype(int)
blocks.append(block)
X = pd.concat(
[X_numeric.reset_index(drop=True)]
+ [b.reset_index(drop=True) for b in blocks],
axis=1,
).fillna(0.0)
meta = {
"feature_names": X.columns.tolist(),
"numeric_features": numeric_features,
"categorical_levels": categorical_levels,
"label_to_int": LABEL_TO_INT,
"int_to_label": INT_TO_LABEL,
}
return X, y, groups, meta
def transform_single(
record: dict | pd.DataFrame,
meta: dict[str, Any],
) -> np.ndarray:
"""Encode a single timestep record for inference."""
if isinstance(record, dict):
df = pd.DataFrame([record.copy()])
else:
df = record.copy()
df = _add_engineered_features(df)
numeric = pd.DataFrame({
col: df.get(col, pd.Series([0.0] * len(df))).astype(float).values
for col in meta["numeric_features"]
})
blocks: list[pd.DataFrame] = [numeric]
for col, levels in meta["categorical_levels"].items():
val = df.get(col, pd.Series([None] * len(df)))
block = pd.get_dummies(
val.astype("category").cat.set_categories(levels),
prefix=col, dummy_na=False,
).astype(int)
for lvl in levels:
cname = f"{col}_{lvl}"
if cname not in block.columns:
block[cname] = 0
block = block[[f"{col}_{lvl}" for lvl in levels]]
blocks.append(block)
X = pd.concat(blocks, axis=1).fillna(0.0)
X = X.reindex(columns=meta["feature_names"], fill_value=0.0)
return X.values.astype(np.float32)
def save_meta(meta: dict[str, Any], path: str | Path) -> None:
serializable = {
"feature_names": meta["feature_names"],
"numeric_features": meta["numeric_features"],
"categorical_levels": meta["categorical_levels"],
"label_to_int": meta["label_to_int"],
"int_to_label": {str(k): v for k, v in meta["int_to_label"].items()},
}
with open(path, "w") as f:
json.dump(serializable, f, indent=2)
def load_meta(path: str | Path) -> dict[str, Any]:
with open(path) as f:
meta = json.load(f)
meta["int_to_label"] = {int(k): v for k, v in meta["int_to_label"].items()}
return meta
if __name__ == "__main__":
import sys
base = Path(sys.argv[1]) if len(sys.argv) > 1 else Path("/mnt/user-data/uploads")
X, y, groups, meta = build_features(base / "malware_samples.csv")
print(f"X shape: {X.shape}")
print(f"y shape: {y.shape}")
print(f"groups: {groups.nunique()} samples")
print(f"n features: {len(meta['feature_names'])}")
print(f"label distribution:\n{y.map(INT_TO_LABEL).value_counts()}")
print(f"X has NaN: {X.isnull().any().any()}")
|