BIS MICRO PROJECT

AN MICRO PROJRCT REPORT ON

FRAUD DETECTION IN TRANSACTIONS

Submitted by

KASHISH MAHOVIA (216300307006)

KARINA SHARMA (226300307120 )
PRUTHA PANCHAL (226300307078)

Student of

Diploma In Engineering
in

computer Department
 Step-by-Step Explanation
 Importing Libraries:

o import pandas as pd

o The code begins by importing the pandas library, which is essential fo data
manipulation and analysis in Python.

 Sample Transaction Data:

data = {

'transaction_id': [1, 2, 3, 4, 5],

'amount': [100, 1500, 200, 3000, 50],

'location': ['USA', 'USA', 'Canada', 'USA', 'Canada'],

'time': ['2023-10-01 10:00', '2023-10-01 10:05', '2023-10-01 10:10', '2023-10-01 10:15', '2023-10-01 10:20'],

'user_id': [101, 101, 102, 101, 102]

}
transaction_id amount location time user_id
1 100 USA 2023-10-01 10:00 101
2 1500 USA 2023-10-01 10:05 101
3 200 Canada 2023-10-01 10:10 102
4 3000 USA 2023-10-01 10:15 101
5 50 Canada 2023-10-01 10:20 102
A dictionary named data is created, containing sample transaction information.
Each key represents a column in the dataset:
o transaction_id: Unique identifier for each transaction.

o amount: The monetary value of the transaction.

o location: The geographical location where the transaction occurred.

o time: The timestamp of the transaction.

o user_id: Identifier for the user who made the transaction.

 CREATING A DATAFRAME:
o transactions = pd.DataFrame(data)

o The data dictionary is converted into a pandas DataFrame

called transactions.

o This structure allows for easy manipulation and analysis of the transaction data.

 DEFINING THRESHOLDS FOR FRAUD DETECTION:

amount_threshold = 2000 # Example threshold for high transaction amount

location_change_threshold = 10 # Example threshold for location changes

suspicious_locations = ['USA', 'Canada'] # Example of known locations

 Several thresholds and criteria are defined for detecting

potential fraud:
o amount_threshold: A threshold of 2000 is set to flag transactions that exceed this
amoun as potentially fraudulent.
o location_change_threshold: This variable is defined but not used in the
current implementation. It could be intended for future enhancements.
o suspicious_locations: A list of known locations (USA and Canada) is defined.
Transactions from locations outside this list may be flagged
 FUNCTION TO DETECT FRAUD:
def detect_fraud(transactions):

fraud_cases = []

o A function named detect_fraud is defined, which takes the transactions

DataFrame as input and initializes an empty list fraud_cases to store any detected fraud
cases.

 ITERATING THROUGH TRANSACTIONS:

o for index, row in transactions.iterrows():

o The function iterates over each transaction in the DataFrame using iterrows(), which returns
both the index and the row data.

 CHECKING FOR HIGH TRANSACTION AMOUNT:

if row['amount'] > amount_threshold:

fraud_cases.append((row['transaction_id'], 'High transaction amount'))

o For each transaction, it checks if the amount exceeds the defined amount_threshold. If it
does, a tuple containing the transaction_id and the reason ('High transaction amount') is
appended to the fraud_cases list.

 CHECKING FOR UNUSUAL LOCATIONS

if row['location'] not in suspicious_locations:

fraud_cases.append((row['transaction_id'], 'Unusual location'))

o The code checks if the location of the transaction is not in

the suspicious_locations list. If it is not, it appends a tuple to fraud_cases indicating
an 'Unusual location'.

 CHECKING FOR MULTIPLE TRANSACTIONS IN A SHORT TIME FRAME

recent_transactions = transactions[(transactions['user_id'] == row['user_id']) &

(transactions['time'] < row['time'])]

if not recent_transactions.empty:
 time_diff = pd.to_datetime(row['time']) - pd.to_datetime(recent_transactions['time'].max())

if time_diff.total_seconds() < 500: # 5 minutes

fraud_cases.append((row['transaction_id'], 'Multiple transactions in short time'))

FRAUD DETECTION IN TRANSACTIONS

Code:

import pandas as pd

# Sample transaction data

data = {

'transaction_id': [1, 2, 3, 4, 5],

'amount': [100, 1500, 200, 3000, 50],

'location': ['USA', 'USA', 'Canada', 'USA', 'Canada'],

'time': ['2023-10-01 10:00', '2023-10-01 10:05', '2023-10-01 10:10', '2023-10-01 10:15', '2023-10-01
10:20'],

'user_id': [101, 101, 102, 101, 102]

# Create a DataFrame

transactions = pd.DataFrame(data)

# Define thresholds for fraud detection

amount_threshold = 1000 # Example threshold for high transaction amount

location_change_threshold = 1 # Example threshold for location changes

suspicious_locations = ['USA', 'Canada'] # Example of known locations

# Function to detect fraud

def detect_fraud(transactions):

fraud_cases = []

for index, row in transactions.iterrows():

 # Check for high transaction amount

if row['amount'] > amount_threshold:

fraud_cases.append((row['transaction_id'], 'High transaction amount'))

# Check for unusual location (assuming user has a known location)

if row['location'] not in suspicious_locations:

fraud_cases.append((row['transaction_id'], 'Unusual location'))

# Check for multiple transactions in a short time frame

recent_transactions = transactions[(transactions['user_id'] == row['user_id']) &

(transactions['time'] < row['time'])]

if not recent_transactions.empty:

time_diff = pd.to_datetime(row['time']) - pd.to_datetime(recent_transactions['time'].max())

if time_diff.total_seconds() < 300: # 5 minutes

fraud_cases.append((row['transaction_id'], 'Multiple transactions in short time'))

return fraud_cases

# Detect fraud in transactions

fraudulent_transactions = detect_fraud(transactions)

# Output results

if fraudulent_transactions:

print("Fraudulent Transactions Detected:")

for transaction in fraudulent_transactions:

print(f"Transaction ID: {transaction[0]}, Reason: {transaction[1]}")

else:

print("No fraudulent transactions detected.")

project output