ML | Credit Card Fraud Detection

Michael Zippo 18.07.2021

The main challenges in detecting credit card fraud are:

Huge data is processed every day, and model building must be fast enough to respond to fraud in a timely manner .
Unbalanced data, i.e. most transactions (99.8%) are not fraudulent, making them difficult to detect.
Data availability because the data is mostly private.
Unclassified data can be another major concern as not every fraudulent transaction is detected and recorded.
Adaptive techniques used against the model by fraudsters.

How to solve these problems?

The model used must be simple and fast enough to detect the anomaly and classify it as a fraudulent transaction as soon as possible .
Imbalances can be dealt with by using some some methods, which we will discuss in the next paragraph.
Data can be scaled down to protect user privacy.
A more reliable source should be taken that double-checks the data at least to train the model.
We can make the model simple and straightforward so that when a cheater adapts to it with just a few tweaks, we can get a new model up and running.

Before getting into the code, he is asked to work on a Jupyter notebook. If not installed on your computer, you can use Google Colab .
You can download the dataset from this link
If the link doesn’t work, go to this link and log in to kaggle to download the dataset.
Code: import all required libraries

# import required packages

import numpy as np

import pandas as pd

import matplotlib.pyplot as plt

import seaborn as sns

from matplotlib import gridspec

Code: data loading

       # Load dataset from CSV file using pan d  
  # the best way is to mount the disk on colaba and  
  # copy the path to the CSV file  
   path   =   "credit.csv"  
   data   =   pd.read_csv (path)

Code: Understanding Data

# Look at the data
data.head ()

Code: data description

# Print data form
# data = data.sample (frac = 0.1, random_state = 48)

print (data.shape)

print (data.describe ())

Output:

 (284807, 31) Time V1 ... Amount Class count 284807.000000 2.848070e + 05 ... 284807.000000 284807.000000  mean  94813.859575 3.919560e-15 ... 88.349619 0.001727 std 47488.145955 1.958696e + 00 ... 250.120109 0.0415 0.000000 -5.640751e + 01 ... 0.000000 0.000000 25% 54201.500000 -9.203734e-01 ... 5.600000 0.000000 50% 84692.000000 1.810880e-02 ... 22.000000 0.000000 75% 139320.500000 1.315642e + 00 ... 77.165000 0.000000 max 172792.000000 2.454930e + 00 ... 25691.160000 1.000000 [8 rows x 31 columns]

Code: data imbalance
Time to explain the data we are dealing with

# Determine the number of fraud cases in the dataset

fraud = data [data [ ’Class’ ] = = 1 ]

valid = data [data [ ’Class’ ] = = 0 ]

outlierFraction = len (fraud) / float ( len (valid))

print (outlierFraction)

print ( ’Fraud Cases: {}’ . format ( len (data [data [ ’Class’ ] = = 1 ])) )

print ( ’Valid Transactions: {} ’ . format ( len (data [data [ ’Class’ ] = = 0 ])))

Only 0.17% fraudulent transactions from all transactions. The data is highly imbalanced. Let’s apply our models first without balancing them, and if we don’t get good accuracy, then we can find a way to balance this dataset. But first, let’s implement the model without it and balance the data only if necessary.

Code: print information about the amount of the fraudulent transaction

print (“Amount details of the fraudulent transaction”)

fraud.Amount .describe ()

Output:

 Amount details of the fraudulent transaction count 492.000000  mean  122.211321 std 256.683288 min 0.000000 25% 1.000000 50% 9.250000 75% 105.890000 max 2125.870000 Name: Amount, dtype: float64

Code: Print amount information for a regular transaction

print (“Detai ls of valid transaction ”)

valid.Amount.describe ()

Output:

 Amount details of valid transaction count 284315.000000  mean  88.291022 std 250.105092 min 0.000000 25 % 5.650000 50% 22.000000 75% 77.050000 max 25691.160000 Name: Amount, dtype: float64

As we can clearly see, the average Money transaction for fraudulent transactions is higher. This makes this problem solvable.

Code: Building a correlation matrix
A correlation matrix graphically gives us an idea of how functions correlate with each other and can help us to predict which features are most relevant for forecasting.

# Correlation matrix

corrmat = data.corr ()

fig = plt.figure (figsize = ( 12 , 9 ))

sns.heatmap (corrmat, vmax = . 8 , square = True )

plt.show ()

In HeatMap we can clearly see that most of the functions are not related to other features, but there are some features that correlate positively or negatively with each other. For example, V2 and V5 correlate strongly negatively with a feature called Amount . We also see some correlation with V20 and Amount . This gives us a deeper understanding of the data available to us.

Code: Split X and Y Values 
Split Data into Input Parameters and Format Output Values

# divide X and Y from dataset

X = data.drop ([ ’ Class’ ], axis = 1 )

Y = data [ "Class " ]

print (X.shape )

print (Y.shape)

# get only values to process
# (this is an empty array with no columns)

xData = X.values

yData = Y.values

Output:

 (284807, 30) (284807,)

Training and bifurcation testing
We will divide the dataset into two main groups. One for training the model and the other for testing the performance of our trained model.

# Using Skicit-learn to split data into training and test cases

from sklearn.model_selection import train_test_split

# Split data into training and test cases

xTrain, xTest, yTrain, yTest = train_test_split (

xData, yData, test_size = 0.2 , random_state = 42 )

Code: Building a random forest model using skicit learn

# Building a RANDOM FOREST classifier

from sklearn.ensemble import RandomForestClassifier

# create a random forest model

rfc = RandomForestClassifier ()

rfc.fit (xTrain, yTrain)
# predictions

yPred = rfc.predict (xTest)

Code: creation of all kinds evaluation parameters

# Classifier score
# print each classifier score
# scored anything

from sklearn.metrics import classification_report, accuracy_score

from sklearn.metrics import precision_score, recall_score

from sklearn.metrics import f1_score, matthews_corrcoef

from sklearn.metrics import con fusion_matrix

n_outliers = len (fraud)

n_errors = (yPred! = yTest). sum ()

print ( "The model used is Random Forest classifier" )

acc = accuracy_score (yTest, yPred)

print ( "The accuracy is {}" . format (ac c))

prec = precision_score (yTest, yPred)

print ( "The precision is {}" . format (prec))

rec = recall_score (yTest, yPred)

print ( "The recall is {}" . format (rec))

f1 = f1_score ( yTest, yPred)

print ( "The F1-Score is {}" . format (f1))

MCC = matthews_corrcoef (yTest, yPred)

print ( "The Matthews correlation coefficient is {}" . format (MCC))

Output:

 The model used is Random Forest classifier The accuracy is 0.9995611109160493 The precision is 0.9866666666666667 The recall is 0.7551020408163265 The F1-Score is 0.8554913294797689 The Matthews correlation coefficient is0.8629589216367891

Code: confusion visualization

/ p>

# print confusion matrix

LABELS = [ ’ Normal’ , ’Fraud’ ]

conf_matrix = confusion_matrix (yTest, yPred)

plt.figure (figsize = ( 12 , 12 ) )

sns.heatmap (conf_matrix, xticklabels = LABELS,

yticklabels = LABELS, ann ot = True , fmt = " d " );

plt.title ( "Confusion matrix" )

plt.ylabel ( ’True class’ )

plt.xlabel ( ’ Predicted class’ )

plt.show ()