-
Notifications
You must be signed in to change notification settings - Fork 0
/
anomaly_detection.py
338 lines (323 loc) · 11.5 KB
/
anomaly_detection.py
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
327
328
329
330
331
332
333
334
335
336
337
338
import time
import numpy as np
# import matplotlib
# import matplotlib.pyplot as plt
import pandas as pd
from sklearn import svm
from nltk.stem.snowball import RussianStemmer
# from sklearn.covariance import EllipticEnvelope
from sklearn.ensemble import IsolationForest
from sklearn.neighbors import LocalOutlierFactor
from sklearn.linear_model import SGDOneClassSVM
from sklearn.kernel_approximation import Nystroem
from sklearn.pipeline import make_pipeline
from sklearn.feature_extraction.text import (
TfidfVectorizer,
CountVectorizer,
HashingVectorizer,
)
from sklearn.preprocessing import (
StandardScaler,
# LabelEncoder,
OneHotEncoder,
PolynomialFeatures,
)
from sklearn.compose import ColumnTransformer
# from sklearn.impute import SimpleImputer
# from sklearn.pipeline import FeatureUnion
# from sklearn.decomposition import PCA, TruncatedSVD
from sklearn.feature_selection import VarianceThreshold
import logging
# logger = logging.getLogger(__name__)
# change for prints in case of docker
print("Starting Exploratory Data Analysis (EDA)")
ueba = pd.read_csv("ueba.csv", index_col=0)
ueba.describe()
print("Cheking unique domains, ids, uid")
print(
f"unique domains = {set(ueba.domain)}, number of unique ids = {len(set(ueba.id))}, number of unique uids {len(set(ueba.uid))}"
)
print("NaN in uid gives us first three anomaly persons:")
anomalies_1 = ueba[ueba.uid.isna()]
print(anomalies_1)
print("Let's count by aggregating lenght of groups ids:")
ueba["len_member_of_groups_ids"] = ueba["member_of_groups_ids"].apply(
lambda x: len(str(x))
)
ueba.groupby(by="len_member_of_groups_ids").count()
print(
"Future investigation in NANs and grops ids gives us some suspicious people with small logins and huge rights:"
)
anomalies_2 = ueba[
np.any(
np.c_[
(ueba.len_member_of_groups_ids == 267).values,
(ueba.len_member_of_groups_ids == 119).values,
# (ueba.len_member_of_groups_ids==115).values,
],
axis=1,
)
]
print(anomalies_2)
print(
"Future investigation in NANs and grops ids gives us some suspicious people with huge rights, we need to check them:"
)
print("Let's prepare data and try some ML algorithms:")
data = ueba.drop(["domain"], axis=1)
data["text"] = data["cn"] + " " + data["title"] + " " + data["who"]
categorical_cols = ["department"]
data.drop(["cn", "title", "who"], axis=1)
data.drop(["id", "uid"], axis=1, inplace=True)
print("Filling NaNs with simple method:")
data[
[
"cn",
"department",
"title",
"who",
"text",
"member_of_indirect_groups_ids",
"member_of_groups_ids",
]
] = data[
[
"cn",
"department",
"title",
"who",
"text",
"member_of_indirect_groups_ids",
"member_of_groups_ids",
]
].fillna(
value="NAN"
)
data.fillna(value="-1", inplace=True)
print("Defining data preprocessing:")
# Some code taken from:
# https://stackoverflow.com/questions/36182502/add-stemming-support-to-countvectorizer-sklearn
# https://scikit-learn.org/stable/auto_examples/miscellaneous/plot_anomaly_comparison.html#sphx-glr-auto-examples-miscellaneous-plot-anomaly-comparison-py
stemmer = RussianStemmer()
analyzer = CountVectorizer().build_analyzer()
def stemmed_words(doc):
return (stemmer.stem(w) for w in analyzer(doc))
# Settings
n_samples = 2633
n_outliers = 5
outliers_fraction = n_outliers / n_samples
n_inliers = n_samples - n_outliers
categorical_cols = ["cn", "department", "title", "who"]
transform_algorithms = [
(
"Features baseline",
make_pipeline(
ColumnTransformer(
[
(
"OHE",
OneHotEncoder(
sparse_output=True, handle_unknown="infrequent_if_exist"
),
categorical_cols,
),
# handle categorical columns as one hot encoding
(
"Text",
HashingVectorizer(
ngram_range=(3, 6), analyzer="char_wb", n_features=20000
),
"text",
),
*[
(
col,
TfidfVectorizer(
ngram_range=(1, 4),
preprocessor=lambda x: " ".join(x.split(";")),
analyzer="word",
),
col,
)
for col in [
"member_of_indirect_groups_ids",
"member_of_groups_ids",
]
], # handle categories as text, interaction up to 4 included
]
),
VarianceThreshold(threshold=0.0001),
PolynomialFeatures(
degree=2, interaction_only=True, include_bias=False
), # intersting interaction features like Is_admin * title
VarianceThreshold(), # remove zero varience features
StandardScaler(
with_mean=False
), # Scaling is important for anomaly detection
),
),
(
"Features without scaling",
make_pipeline(
ColumnTransformer(
[
(
"OHE",
OneHotEncoder(
sparse_output=True, handle_unknown="infrequent_if_exist"
),
categorical_cols,
),
# handle categorical columns as one hot encoding
(
"Text",
HashingVectorizer(
ngram_range=(3, 6), analyzer="char_wb", n_features=20000
),
"text",
),
*[
(
col,
TfidfVectorizer(
ngram_range=(1, 4),
preprocessor=lambda x: " ".join(x.split(";")),
analyzer="word",
),
col,
)
for col in [
"member_of_indirect_groups_ids",
"member_of_groups_ids",
]
], # handle categories as text, interaction up to 4 included
]
),
VarianceThreshold(threshold=0.0001),
PolynomialFeatures(
degree=2, interaction_only=True, include_bias=False
), # intersting interaction features like Is_admin * title
VarianceThreshold(), # remove zero varience features
),
),
(
"Features CountVectorizer with Stemmer",
make_pipeline(
ColumnTransformer(
[
(
"OHE",
OneHotEncoder(
sparse_output=True, handle_unknown="infrequent_if_exist"
),
categorical_cols,
),
# handle categorical columns as one hot encoding
(
"Text",
CountVectorizer(analyzer=stemmed_words),
"text",
),
*[
(
col,
TfidfVectorizer(
ngram_range=(1, 4),
preprocessor=lambda x: " ".join(x.split(";")),
analyzer="word",
),
col,
)
for col in [
"member_of_indirect_groups_ids",
"member_of_groups_ids",
]
], # handle categories as text, interaction up to 4 included
]
),
VarianceThreshold(threshold=0.001),
PolynomialFeatures(
degree=2, interaction_only=True, include_bias=False
), # intersting interaction features like Is_admin * title
VarianceThreshold(), # remove zero varience features
StandardScaler(
with_mean=False
), # Scaling is important for anomaly detection
),
),
(
"Features TfidfVectorizer with Stemmer",
make_pipeline(
ColumnTransformer(
[
(
"OHE",
OneHotEncoder(
sparse_output=True, handle_unknown="infrequent_if_exist"
),
categorical_cols,
),
# handle categorical columns as one hot encoding
(
"Text",
TfidfVectorizer(analyzer=stemmed_words),
"text",
),
*[
(
col,
TfidfVectorizer(
ngram_range=(1, 4),
preprocessor=lambda x: " ".join(x.split(";")),
analyzer="word",
),
col,
)
for col in [
"member_of_indirect_groups_ids",
"member_of_groups_ids",
]
], # handle categories as text, interaction up to 4 included
]
),
VarianceThreshold(threshold=0.001),
PolynomialFeatures(
degree=2, interaction_only=True, include_bias=False
), # intersting interaction features like Is_admin * title
VarianceThreshold(), # remove zero varience features
StandardScaler(
with_mean=False
), # Scaling is important for anomaly detection
),
),
]
# Define algorithms:
anomaly_algorithms = [
# ("Robust covariance", EllipticEnvelope(contamination=outliers_fraction)),
("One-Class SVM", svm.OneClassSVM(nu=outliers_fraction, kernel="rbf", gamma=0.1)),
(
"Isolation Forest",
IsolationForest(contamination=outliers_fraction, random_state=42),
),
# (
# "Local Outlier Factor",
# LocalOutlierFactor(n_neighbors=10, contamination=outliers_fraction),
# ),
]
rng = np.random.RandomState(42)
for i_pipe, (pipe_name, pipeline) in enumerate(transform_algorithms):
for name, algorithm in anomaly_algorithms:
t0 = time.time()
detector = make_pipeline(pipeline, algorithm)
detector.fit(data)
detector
t1 = time.time()
print(
"Elapsed time for pipeline {} and algorithm {}: {:.2f}s".format(
pipe_name, name, t1 - t0
)
)
y_pred = detector.predict(data)
print(f"Number of outliers: {sum(y_pred == 0)}")
print("No new outliers with ML")
anomalies = pd.concat([anomalies_1, anomalies_2], axis=0).drop(1616, axis=0)
anomalies.to_csv("anomalies.csv")