def agglomerative(normal, anormal, connectivity, n_clusters=2, linkage='ward'):
X = pd.concat([normal, anormal], axis=0)
X_fraude = X[['id_siniestro', 'FRAUDE']]
del X['FRAUDE']
del X['id_siniestro']
db = AgglomerativeClustering(n_clusters=n_clusters, linkage=linkage, connectivity=connectivity).fit(X)
labels = db.labels_
labels_df = pd.DataFrame(labels, index = X.index, columns=['Clusters'])
labels = db.labels_
labels_unique = np.unique(labels)
n_clusters_ = len(labels_unique)
print(n_clusters_)
comparative = pd.concat([X_fraude, labels_df], axis=1)
f1, f2, fscore, df_clusters = fs.fraud_score(comparative.drop(['id_siniestro'], axis=1), 'FRAUDE', 'Clusters')
comparative['FRAUDE_Clusters'] = pd.Series(0, index=comparative.index)
comparative['FRAUDE'] = comparative['FRAUDE'].map(int)
comparative.loc[comparative['FRAUDE'] == 1, 'FRAUDE_Clusters'] = 1
comparative.loc[comparative['Clusters'].isin(df_clusters), 'FRAUDE_Clusters'] = 1
return f1, f2, fscore, comparative
def isolation_forest(normal, anormal, contamination=0.1, n_estimators=50):
X = pd.concat([normal, anormal], axis=0)
X_fraude = X[['id_siniestro', 'FRAUDE']]
del X['FRAUDE']
del X['id_siniestro']
db = IsolationForest(n_estimators = n_estimators, max_samples=X.shape[0],
bootstrap=True, verbose=1, random_state=42,
contamination=contamination)
db.fit(X)
labels = db.predict(X)
labels_df = pd.DataFrame(labels, index=X.index, columns=['Clusters'])
labels_unique = np.unique(labels)
n_clusters_ = len(labels_unique)
comparative = pd.concat([X_fraude, labels_df], axis=1)
f1, f2, fscore, df_clusters = fs.fraud_score(comparative.drop(['id_siniestro'], axis=1), 'FRAUDE', 'Clusters')
comparative['FRAUDE_Clusters'] = pd.Series(0, index=comparative.index)
comparative['FRAUDE'] = comparative['FRAUDE'].map(int)
comparative.loc[comparative['FRAUDE'] == 1, 'FRAUDE_Clusters'] = 1
comparative.loc[comparative['Clusters'].isin(df_clusters), 'FRAUDE_Clusters'] = 1
return f1, f2, fscore, comparative
def super_vector(normal, anormal, nu=0.1, gamma=0.1):
X = pd.concat([normal, anormal], axis=0)
X_fraude = X[['id_siniestro', 'FRAUDE']]
del X['FRAUDE']
del X['id_siniestro']
db = svm.OneClassSVM(nu=nu, kernel="linear", gamma=gamma)
db.fit(X)
labels = db.predict(X)
labels_df = pd.DataFrame(labels, index=X.index, columns=['Clusters'])
labels_unique = np.unique(labels)
n_clusters_ = len(labels_unique)
comparative = pd.concat([X_fraude, labels_df], axis=1)
f1, f2, fscore, df_clusters = fs.fraud_score(
comparative.drop(['id_siniestro'], axis=1), 'FRAUDE', 'Clusters')
comparative['FRAUDE_Clusters'] = pd.Series(0, index=comparative.index)
comparative['FRAUDE'] = comparative['FRAUDE'].map(int)
comparative.loc[comparative['FRAUDE'] == 1, 'FRAUDE_Clusters'] = 1
comparative.loc[comparative['Clusters'].isin(df_clusters),
'FRAUDE_Clusters'] = 1
return f1, f2, fscore, comparative
def dbscan(normal, anormal, eps=0.3, min_samples=175, leaf_size=30):
X = pd.concat([normal, anormal], axis=0)
X_fraude = X[['id_siniestro', 'FRAUDE']]
del X['FRAUDE']
del X['id_siniestro']
db = DBSCAN(eps=eps,
min_samples=min_samples,
leaf_size=leaf_size,
n_jobs=-1).fit(X)
labels = db.labels_
labels_df = pd.DataFrame(labels, index=X.index, columns=['Clusters'])
labels = db.labels_
labels_unique = np.unique(labels)
n_clusters_ = len(labels_unique)
print(n_clusters_)
comparative = pd.concat([X_fraude, labels_df], axis=1)
f1, f2, fscore, df_clusters = fs.fraud_score(
comparative.drop(['id_siniestro'], axis=1), 'FRAUDE', 'Clusters')
comparative['FRAUDE_Clusters'] = pd.Series(0, index=comparative.index)
comparative['FRAUDE'] = comparative['FRAUDE'].map(int)
comparative.loc[comparative['FRAUDE'] == 1, 'FRAUDE_Clusters'] = 1
comparative.loc[comparative['Clusters'].isin(df_clusters),
'FRAUDE_Clusters'] = 1
return f1, f2, fscore, comparative
def hdbscan(normal, anormal, min_cluster_size=15, min_samples=5, csm='eom'):
X = pd.concat([normal, anormal], axis=0)
X_fraude = X[['FRAUDE']]
del X['FRAUDE']
db = HDBSCAN(min_cluster_size=min_cluster_size,
min_samples=min_samples,
cluster_selection_method=csm,
allow_single_cluster=True,
core_dist_n_jobs=-1).fit(X)
labels = db.labels_
labels_df = pd.DataFrame(labels, index=X.index, columns=['Clusters'])
labels = db.labels_
labels_unique = np.unique(labels)
n_clusters_ = len(labels_unique)
print(n_clusters_)
comparative = pd.concat([X_fraude, labels_df], axis=1)
f1, f2, fscore, df_clusters = fs.fraud_score(
comparative.drop(['id_siniestro'], axis=1), 'FRAUDE', 'Clusters')
comparative['FRAUDE_Clusters'] = pd.Series(0, index=comparative.index)
comparative['FRAUDE'] = comparative['FRAUDE'].map(int)
comparative.loc[comparative['FRAUDE'] == 1, 'FRAUDE_Clusters'] = 1
comparative.loc[comparative['Clusters'].isin(df_clusters),
'FRAUDE_Clusters'] = 1
return f1, f2, fscore, comparative
def mean_shift(normal, anormal, quantile=0.5):
X = pd.concat([normal, anormal], axis=0)
X_fraude = X[['id_siniestro', 'FRAUDE']]
del X['FRAUDE']
del X['id_siniestro']
bandwith = estimate_bandwidth(X.values, quantile=quantile, random_state=42)
db = MeanShift(bandwidth=bandwith,
bin_seeding=True,
cluster_all=False,
min_bin_freq=50,
n_jobs=-1).fit(X)
labels = db.labels_
labels_df = pd.DataFrame(labels, index=X.index, columns=['Clusters'])
labels = db.labels_
labels_unique = np.unique(labels)
n_clusters_ = len(labels_unique)
cluster_centers = db.cluster_centers_
comparative = pd.concat([X_fraude, labels_df], axis=1)
f1, f2, fscore, df_clusters = fs.fraud_score(
comparative.drop(['id_siniestro'], axis=1), 'FRAUDE', 'Clusters')
comparative['FRAUDE_Clusters'] = pd.Series(0, index=comparative.index)
comparative['FRAUDE'] = comparative['FRAUDE'].map(int)
comparative.loc[comparative['FRAUDE'] == 1, 'FRAUDE_Clusters'] = 1
comparative.loc[comparative['Clusters'].isin(df_clusters),
'FRAUDE_Clusters'] = 1
return f1, f2, fscore, comparative
def mini_batch_kmeans(normal,
anormal,
n_clusters=2,
max_iter=100,
batch_size=100):
X = pd.concat([normal, anormal], axis=0)
X_fraude = X[['id_siniestro', 'FRAUDE']]
del X['FRAUDE']
del X['id_siniestro']
db = MiniBatchKMeans(n_clusters=n_clusters,
max_iter=max_iter,
batch_size=batch_size,
random_state=541)
db.fit(X)
labels = db.predict(X)
labels_df = pd.DataFrame(labels, index=X.index, columns=['Clusters'])
labels_unique = np.unique(labels)
n_clusters_ = len(labels_unique)
comparative = pd.concat([X_fraude, labels_df], axis=1)
f1, f2, fscore, df_clusters = fs.fraud_score(
comparative.drop(['id_siniestro'], axis=1), 'FRAUDE', 'Clusters')
comparative['FRAUDE_Clusters'] = pd.Series(0, index=comparative.index)
comparative['FRAUDE'] = comparative['FRAUDE'].map(int)
comparative.loc[comparative['FRAUDE'] == 1, 'FRAUDE_Clusters'] = 1
comparative.loc[comparative['Clusters'].isin(df_clusters),
'FRAUDE_Clusters'] = 1
return f1, f2, fscore, comparative
def gaussian_mixture(normal,
anormal,
n_components,
cov,
tol,
max_iter,
reg_covar,
type='Gaussian'):
X = pd.concat([normal, anormal], axis=0)
X_fraude = X[['id_siniestro', 'FRAUDE']]
del X['FRAUDE']
del X['id_siniestro']
if type == 'Gaussian':
db = mixture.GaussianMixture(n_components=n_components,
covariance_type=cov,
tol=tol,
max_iter=max_iter,
init_params='kmeans',
random_state=541,
reg_covar=reg_covar).fit(X)
if type == 'Bayesian':
db = mixture.BayesianGaussianMixture(n_components=n_components,
covariance_type=cov,
tol=tol,
max_iter=max_iter,
init_params='kmeans',
random_state=541).fit(X)
labels = db.predict(X)
labels_df = pd.DataFrame(labels, index=X.index, columns=['Clusters'])
labels_unique = np.unique(labels)
n_clusters_ = len(labels_unique)
print(n_clusters_)
comparative = pd.concat([X_fraude, labels_df], axis=1)
f1, f2, fscore, df_clusters = fs.fraud_score(
comparative.drop(['id_siniestro'], axis=1), 'FRAUDE', 'Clusters')
comparative['FRAUDE_Clusters'] = pd.Series(0, index=comparative.index)
comparative['FRAUDE'] = comparative['FRAUDE'].map(int)
comparative.loc[comparative['FRAUDE'] == 1, 'FRAUDE_Clusters'] = 1
comparative.loc[comparative['Clusters'].isin(df_clusters),
'FRAUDE_Clusters'] = 1
return f1, f2, fscore, comparative