발표2

Author

김보람

Published

December 16, 2023

imports

import numpy as np
import pandas as pd
import matplotlib.pyplot as plt 
import networkx as nx
import sklearn
import xgboost as xgb
import catboost as cb

# sklearn
from sklearn import model_selection # split함수이용
from sklearn import ensemble # RF,GBM
from sklearn import metrics
from sklearn.metrics import precision_score, recall_score, f1_score
from sklearn.svm import SVC
from sklearn.ensemble import RandomForestClassifier
from sklearn.naive_bayes import GaussianNB

# gnn
import torch
import torch.nn.functional as F
import torch_geometric
from torch_geometric.nn import GCNConv


# 불균형 데이터 처리
def down_sample_textbook(df):
    df_majority = df[df.is_fraud==0].copy()
    df_minority = df[df.is_fraud==1].copy()
    df_maj_dowsampled = sklearn.utils.resample(df_majority, n_samples=len(df_minority), replace=False, random_state=42)
    df_downsampled = pd.concat([df_minority, df_maj_dowsampled])
    return df_downsampled


# 고객별 거래 시간 차이 
def compute_time_difference(group):
    n = len(group)
    result = []
    for i in range(n):
        for j in range(n):
            time_difference = abs(group.iloc[i].trans_date_trans_time.value - group.iloc[j].trans_date_trans_time.value)
            result.append([group.iloc[i].name, group.iloc[j].name, time_difference])
    return result

# GCN_model_train/test 
def mask(df):
    df_tr,df_test = sklearn.model_selection.train_test_split(df, random_state=42)
    N = len(df)
    train_mask = [i in df_tr.index for i in range(N)]
    test_mask = [i in df_test.index for i in range(N)]
    train_mask = np.array(train_mask)
    test_mask = np.array(test_mask)
    return train_mask, test_mask


# wieght 값 설정-> edge_index 적용
def edge_index_selected(edge_index):
    theta = edge_index[:,2].mean()
    edge_index[:,2] = (np.exp(-edge_index[:,2]/theta) != 1)*(np.exp(-edge_index[:,2]/theta))
    edge_index = edge_index.tolist()
    mean_ = np.array(edge_index)[:,2].mean()
    selected_edges = [(int(row[0]), int(row[1])) for row in edge_index if row[2] > mean_]
    edge_index_selected = torch.tensor(selected_edges, dtype=torch.long).t()
    return edge_index_selected

데이터

fraudTrain = pd.read_csv("~/Desktop/fraudTrain.csv").iloc[:,1:]
fraudTrain = fraudTrain.assign(trans_date_trans_time= list(map(lambda x: pd.to_datetime(x), fraudTrain.trans_date_trans_time)))
fraudTrain
trans_date_trans_time cc_num merchant category amt first last gender street city ... lat long city_pop job dob trans_num unix_time merch_lat merch_long is_fraud
0 2019-01-01 00:00:00 2.703190e+15 fraud_Rippin, Kub and Mann misc_net 4.97 Jennifer Banks F 561 Perry Cove Moravian Falls ... 36.0788 -81.1781 3495 Psychologist, counselling 1988-03-09 0b242abb623afc578575680df30655b9 1325376018 36.011293 -82.048315 0
1 2019-01-01 00:00:00 6.304230e+11 fraud_Heller, Gutmann and Zieme grocery_pos 107.23 Stephanie Gill F 43039 Riley Greens Suite 393 Orient ... 48.8878 -118.2105 149 Special educational needs teacher 1978-06-21 1f76529f8574734946361c461b024d99 1325376044 49.159047 -118.186462 0
2 2019-01-01 00:00:00 3.885950e+13 fraud_Lind-Buckridge entertainment 220.11 Edward Sanchez M 594 White Dale Suite 530 Malad City ... 42.1808 -112.2620 4154 Nature conservation officer 1962-01-19 a1a22d70485983eac12b5b88dad1cf95 1325376051 43.150704 -112.154481 0
3 2019-01-01 00:01:00 3.534090e+15 fraud_Kutch, Hermiston and Farrell gas_transport 45.00 Jeremy White M 9443 Cynthia Court Apt. 038 Boulder ... 46.2306 -112.1138 1939 Patent attorney 1967-01-12 6b849c168bdad6f867558c3793159a81 1325376076 47.034331 -112.561071 0
4 2019-01-01 00:03:00 3.755340e+14 fraud_Keeling-Crist misc_pos 41.96 Tyler Garcia M 408 Bradley Rest Doe Hill ... 38.4207 -79.4629 99 Dance movement psychotherapist 1986-03-28 a41d7549acf90789359a9aa5346dcb46 1325376186 38.674999 -78.632459 0
... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ...
1048570 2020-03-10 16:07:00 6.011980e+15 fraud_Fadel Inc health_fitness 77.00 Haley Wagner F 05561 Farrell Crescent Annapolis ... 39.0305 -76.5515 92106 Accountant, chartered certified 1943-05-28 45ecd198c65e81e597db22e8d2ef7361 1362931649 38.779464 -76.317042 0
1048571 2020-03-10 16:07:00 4.839040e+15 fraud_Cremin, Hamill and Reichel misc_pos 116.94 Meredith Campbell F 043 Hanson Turnpike Hedrick ... 41.1826 -92.3097 1583 Geochemist 1999-06-28 c00ce51c6ebb7657474a77b9e0b51f34 1362931670 41.400318 -92.726724 0
1048572 2020-03-10 16:08:00 5.718440e+11 fraud_O'Connell, Botsford and Hand home 21.27 Susan Mills F 005 Cody Estates Louisville ... 38.2507 -85.7476 736284 Engineering geologist 1952-04-02 17c9dc8b2a6449ca2473726346e58e6c 1362931711 37.293339 -84.798122 0
1048573 2020-03-10 16:08:00 4.646850e+18 fraud_Thompson-Gleason health_fitness 9.52 Julia Bell F 576 House Crossroad West Sayville ... 40.7320 -73.1000 4056 Film/video editor 1990-06-25 5ca650881b48a6a38754f841c23b77ab 1362931718 39.773077 -72.213209 0
1048574 2020-03-10 16:08:00 2.283740e+15 fraud_Buckridge PLC misc_pos 6.81 Shannon Williams F 9345 Spencer Junctions Suite 183 Alpharetta ... 34.0770 -84.3033 165556 Prison officer 1997-12-27 8d0a575fe635bbde12f1a2bffc126731 1362931730 33.601468 -83.891921 0

1048575 rows × 22 columns

데이터정리

_df1 = fraudTrain[fraudTrain["is_fraud"] == 0].sample(frac=0.20, random_state=42)
_df2 = fraudTrain[fraudTrain["is_fraud"] == 1]
df02 = pd.concat([_df1,_df2])
df02.shape
(214520, 22)
df50 = down_sample_textbook(df02)
df50 = df50.reset_index()
df50.shape
(12012, 23)

tr/test

mask(df50)
(array([False,  True,  True, ...,  True, False,  True]),
 array([ True, False, False, ..., False,  True, False]))
train_mask, test_mask = mask(df50)
df50_tr,df50_test = sklearn.model_selection.train_test_split(df50, random_state=42)

edge_index 설정

edge_index = np.load('edge_index_list_plus50.npy').astype(np.float64)
edge_index.shape
(200706, 3)
edge_index
array([[1.023000e+03, 1.023000e+03, 0.000000e+00],
       [1.023000e+03, 1.024000e+03, 4.200000e+12],
       [1.023000e+03, 1.028000e+03, 7.764000e+13],
       ...,
       [1.194400e+04, 9.782000e+03, 9.528000e+13],
       [1.194400e+04, 1.176700e+04, 3.055758e+16],
       [1.194400e+04, 1.194400e+04, 0.000000e+00]])
theta = edge_index[:,2].mean()
edge_index[:,2] = (np.exp(-edge_index[:,2]/theta) != 1)*(np.exp(-edge_index[:,2]/theta))
edge_index = edge_index.tolist()
selected_edges = [(int(row[0]), int(row[1])) for row in edge_index if row[2]!= 0]
selected_edges = np.array(selected_edges)
selected_edges.shape
(188590, 2)
a = selected_edges.reshape(2,-1)
a.shape
(2, 188590)
a = torch.tensor(a)

data설정(x, edge_index, y)

x = torch.tensor(df50['amt'], dtype=torch.float).reshape(-1,1)
y = torch.tensor(df50['is_fraud'],dtype=torch.int64)
data = torch_geometric.data.Data(x=x, edge_index = a, y=y, train_mask = train_mask, test_mask = test_mask)
data
Data(x=[12012, 1], edge_index=[2, 188590], y=[12012], train_mask=[12012], test_mask=[12012])

모델 적합

분석 1(GCN)

torch.manual_seed(202250926)

class GCN1(torch.nn.Module):
    def __init__(self):
        super().__init__()
        self.conv1 = GCNConv(1, 32)
        self.conv2 = GCNConv(32,2)

    def forward(self, data):
        x, edge_index = data.x, data.edge_index

        x = self.conv1(x, edge_index)
        x = F.relu(x)
        x = F.dropout(x, training=self.training)
        x = self.conv2(x, edge_index)

        return F.log_softmax(x, dim=1)

X = (data.x[data.train_mask]).numpy()
XX = (data.x[data.test_mask]).numpy()
y = (data.y[data.train_mask]).numpy()
yy = (data.y[data.test_mask]).numpy()

model = GCN1()
optimizer = torch.optim.Adam(model.parameters(), lr=0.01, weight_decay=5e-4)
model.train()
for epoch in range(400):
    optimizer.zero_grad()
    out = model(data)
    loss = F.nll_loss(out[data.train_mask], data.y[data.train_mask])
    loss.backward()
    optimizer.step()
model.eval()

pred = model(data).argmax(dim=1)
yyhat = pred[data.test_mask]


metrics = [sklearn.metrics.accuracy_score,
           sklearn.metrics.precision_score,
           sklearn.metrics.recall_score,
           sklearn.metrics.f1_score]


_results1= pd.DataFrame({m.__name__:[m(yy,yyhat).round(6)] for m in metrics},index=['분석1'])
_results1
accuracy_score precision_score recall_score f1_score
분석1 0.552115 0.589397 0.373764 0.457443

분석2(로지스틱 회귀)

torch.manual_seed(202250926)
X = np.array(df50_tr.loc[:,['amt']])
XX = np.array(df50_test.loc[:,['amt']])
y = np.array(df50_tr.is_fraud)
yy = np.array(df50_test.is_fraud)

lrnr = sklearn.linear_model.LogisticRegression()

lrnr.fit(X,y)

#thresh = y.mean()
#yyhat = (lrnr.predict_proba(XX)> thresh)[:,-1]
yyhat = lrnr.predict(XX) 

yyhat

metrics = [sklearn.metrics.accuracy_score,
           sklearn.metrics.precision_score,
           sklearn.metrics.recall_score,
           sklearn.metrics.f1_score]

_results2= pd.DataFrame({m.__name__:[m(yy,yyhat).round(6)] for m in metrics},index=['분석2'])
_results2
accuracy_score precision_score recall_score f1_score
분석2 0.849484 0.933279 0.756098 0.835397

분석3(XGBoost)

import xgboost as xgb


torch.manual_seed(202250926)
X = np.array(df50_tr.loc[:, ['amt']])
XX = np.array(df50_test.loc[:, ['amt']])
y = np.array(df50_tr.is_fraud)
yy = np.array(df50_test.is_fraud)

lrnr = xgb.XGBClassifier()
 
lrnr.fit(X,y)
yyhat = lrnr.predict(XX)


metrics = [sklearn.metrics.accuracy_score,
           sklearn.metrics.precision_score,
           sklearn.metrics.recall_score,
           sklearn.metrics.f1_score]

_results3= pd.DataFrame({m.__name__:[m(yy,yyhat).round(6)] for m in metrics},index=['분석3'])
_results3
accuracy_score precision_score recall_score f1_score
분석3 0.88012 0.886957 0.874094 0.880478

분석4(Light GBM)

import lightgbm as lgb

torch.manual_seed(202250926)
X = np.array(df50_tr.loc[:, ['amt']])
XX = np.array(df50_test.loc[:, ['amt']])
y = np.array(df50_tr.is_fraud)
yy = np.array(df50_test.is_fraud)


lrnr = lgb.LGBMClassifier()

lrnr.fit(X, y)
yyhat = lrnr.predict(XX)

metrics = [sklearn.metrics.accuracy_score,
           sklearn.metrics.precision_score,
           sklearn.metrics.recall_score,
           sklearn.metrics.f1_score]

_results4 = pd.DataFrame({m.__name__: [m(yy, yyhat).round(6)] for m in metrics}, index=['분석4'])
_results4
accuracy_score precision_score recall_score f1_score
분석4 0.885115 0.893817 0.87673 0.885191

분석5(CatBoost)


torch.manual_seed(202250926)
X = np.array(df50_tr.loc[:, ['amt']])
XX = np.array(df50_test.loc[:, ['amt']])
y = np.array(df50_tr.is_fraud)
yy = np.array(df50_test.is_fraud)

lrnr = cb.CatBoostClassifier(verbose=False)

lrnr.fit(X, y)

yyhat = lrnr.predict(XX)


metrics = [sklearn.metrics.accuracy_score,
           sklearn.metrics.precision_score,
           sklearn.metrics.recall_score,
           sklearn.metrics.f1_score]


_results5 = pd.DataFrame({m.__name__: [m(yy, yyhat).round(6)] for m in metrics}, index=['분석5'])
_results5
accuracy_score precision_score recall_score f1_score
분석5 0.886447 0.895161 0.878049 0.886522

분석6(One class SVM)

from sklearn.svm import OneClassSVM

torch.manual_seed(202250926)

X = np.array(df50_tr.loc[:, ['amt']])
XX = np.array(df50_test.loc[:, ['amt']])
y = np.array(df50_tr.is_fraud)
yy = np.array(df50_test.is_fraud)

ocsvm = OneClassSVM(gamma='auto')  
ocsvm.fit(X[y == 0])  


yyhat_ocsvm = ocsvm.predict(XX)

yyhat_ocsvm_binary = np.where(yyhat_ocsvm == 1, 0, 1)



metrics_ocsvm = [
    sklearn.metrics.accuracy_score,
    sklearn.metrics.precision_score,
    sklearn.metrics.recall_score,
    sklearn.metrics.f1_score
]

_results6 = pd.DataFrame({m.__name__: [m(yy, yyhat_ocsvm_binary).round(6)] for m in metrics_ocsvm}, index=['분석6'])
_results6
accuracy_score precision_score recall_score f1_score
분석6 0.658342 0.614559 0.868161 0.719672

svm_df02


df02_tr,df02_test = sklearn.model_selection.train_test_split(df02, random_state=42)
from sklearn.svm import OneClassSVM

torch.manual_seed(202250926)

X = np.array(df02_tr.loc[:, ['amt']])
XX = np.array(df02_test.loc[:, ['amt']])
y = np.array(df02_tr.is_fraud)
yy = np.array(df02_test.is_fraud)

ocsvm = OneClassSVM(gamma='auto')  
ocsvm.fit(X[y == 0])  


yyhat_ocsvm = ocsvm.predict(XX)

yyhat_ocsvm_binary = np.where(yyhat_ocsvm == 1, 0, 1)



metrics_ocsvm = [
    sklearn.metrics.accuracy_score,
    sklearn.metrics.precision_score,
    sklearn.metrics.recall_score,
    sklearn.metrics.f1_score
]

_results7 = pd.DataFrame({m.__name__: [m(yy, yyhat_ocsvm_binary).round(6)] for m in metrics_ocsvm}, index=['분석7'])
_results7
accuracy_score precision_score recall_score f1_score
분석7 0.505314 0.050497 0.892994 0.095589

svm_fraudtrain


fr_tr,fr_test = sklearn.model_selection.train_test_split(fraudTrain, random_state=42)
from sklearn.svm import OneClassSVM

torch.manual_seed(202250926)

X = np.array(fr_tr.loc[:, ['amt']])
XX = np.array(fr_test.loc[:, ['amt']])
y = np.array(fr_tr.is_fraud)
yy = np.array(fr_test.is_fraud)

ocsvm = OneClassSVM(gamma='auto')  
ocsvm.fit(X[y == 0])  


yyhat_ocsvm = ocsvm.predict(XX)

yyhat_ocsvm_binary = np.where(yyhat_ocsvm == 1, 0, 1)



metrics_ocsvm = [
    sklearn.metrics.accuracy_score,
    sklearn.metrics.precision_score,
    sklearn.metrics.recall_score,
    sklearn.metrics.f1_score
]

_results8 = pd.DataFrame({m.__name__: [m(yy, yyhat_ocsvm_binary).round(6)] for m in metrics_ocsvm}, index=['분석8'])
_results8

정리

구분 Train Test 모형 설명변수 비고
분석1 df50_tr df50_test Proposed amt
분석2 df50_tr df50_test 로지스틱 회귀 amt
분석3 df50_tr df50_test XGBoost amt
분석4 df50_tr df50_test LightGBM amt
분석5 df50_tr df50_test CatBoost amt
분석6 df50_tr df50_test oneSVM amt 
lst = [_results1, _results2,_results3,_results4, _results5, _results6]
pd.concat(lst)
accuracy_score precision_score recall_score f1_score
분석1 0.902098 0.862478 0.959130 0.908240
분석2 0.849484 0.933279 0.756098 0.835397
분석3 0.880120 0.886957 0.874094 0.880478
분석4 0.885115 0.893817 0.876730 0.885191
분석5 0.886447 0.895161 0.878049 0.886522
분석6 0.658342 0.614559 0.868161 0.719672

끄적끄적

-

\(A\): 연결 정보

\(\hat A\): 연결 정보에 자기 자신의 노드 추가

\(D\): \(A\)를 표준화하기 위한 matrix

\(\hat D^{-1/2} \hat A \hat D^{-1/2}\): 연결정보에 대한 matrix

\(\hat D^{-1/2} \hat A \hat D^{-1/2} X\): \(X\)를 평행이동한 느낌

\(\Theta\): weight를 곱하는 과정

-

\(X\): \(N\)행의 행렬

\(y\): 길이 \(N\)의 벡터, 사기거래 레이블

\(\cal I\): cc_num의 집합

\(T_i, i \in \cal I\): 고객의 거래 시간

\({\cal D} := \{ (X_{i,t}, y_{i,t}): i \in {\cal I}, t \in \cal T_i \}\)

\({\cal V} = \{ v_{i,t}: i \in \cal I, t \in \cal T_i \}\)

\(|{\cal V}| = \sum_{i \in \cal I} |T_i| = \cal N\)

\({\epsilon} = \cup_{i \in {\cal I}} \{ ( v_{i,t}, v_{i,s}) : t,s \in T_{i}, t\neq s\}\)

\(W_i : exp( \frac{-{|t-s|}_2^2}{\theta})\)

\(W: N \times N\) matrix