Anomaly Detection with OneClassSVM and Isolation Forest

anomaly-detection.py

import numpy as np
import pandas as pd
from sklearn.svm import OneClassSVM
from sklearn.ensemble import IsolationForest
from sklearn.model_selection import KFold
from sklearn.metrics import accuracy_score

def df2np(df, dtype=np.float64):
    return df.to_numpy().astype(dtype)

limit_samples = None

kf = KFold()

# classifier = OneClassSVM()
classifier = IsolationForest()
df = pd.read_csv("features.csv")

# filter NA values
df = df[df['det_conf'].notnull()]

# split given label
ok_df = df[df['label']=="ok"]
nok_df = df[df['label']=="nok"]

# select only features columns
feature_columns = ['det_width', 'det_height', 'det_ar', 'det_conf',
    'det_lm_0_x', 'det_lm_1_x', 'det_lm_2_x', 'det_lm_3_x', 'det_lm_4_x',
    'det_lm_0_y', 'det_lm_1_y', 'det_lm_2_y', 'det_lm_3_y', 'det_lm_4_y']

ok_df = ok_df[feature_columns]
nok_df = nok_df[feature_columns]

# build numpy array
X_ok = df2np(ok_df)[:limit_samples]
X_nok = df2np(nok_df)[:limit_samples]

# cross-validation
ok_accs = []
nok_accs = []
for train_index, test_index in kf.split(X_ok):
    X_train, X_test = X_ok[train_index], X_ok[test_index]

    classifier.fit(X_train)

    ok_preds = classifier.predict(X_test)
    ok_accs.append(accuracy_score([1 for _ in ok_preds], ok_preds))

    nok_preds = classifier.predict(X_nok)
    nok_accs.append(accuracy_score([-1 for _ in nok_preds], nok_preds))

print(f"OK Accuracy: {np.mean(ok_accs), np.std(ok_accs)}")
print(f"NOK Accuracy: {np.mean(nok_accs), np.std(nok_accs)}")