Credit-Card-Fraud-Detection-

1 Introduction

1.1 Aim and Objectives

The primary aim of this study is to analyse a credit card fraud dataset to the identify patterns and gain a comprehensive insight that can aid in the detection and prevention of fraudulent activities. This will assist the banking industry in making critical step by which their customers use their credit card for transactions.
The transaction data set that is being explored will also assist the company in exploring their customer spending habits, age groups that transact the most, time period where transactions occur, helping the company plan resource allocation there by optimising response time from their customer care departments (Author, 2024). The objectives that will be carried out in this study are as follows:

1. Perform exploratory data analysis on the fraud dataset to understand the nature and characteristics of the data.
2. Identify key factors and variables that contribute to the likelihood of fraudulent transactions.
3. Develop strategies and techniques for detecting and preventing fraud based on the insights gained from the analysis.
4. Compare the effectiveness and efficiency of SAS and Tableau in performing fraud analytics and data visualization.

   1.2 Background Information

   Fraud is a pervasive issue that affects businesses across various industries, leading to significant financial losses and reputational damage. With the advent of digital transactions and e-commerce, the risk of fraud has escalated, necessitating robust fraud detection and prevention mechanisms. This study aims to leverage data analytics and historical transaction data to uncover insights and develop robust fraud prevention strategies and anomalies that can indicate fraudulent activities, thereby enabling organizations to take proactive measures to mitigate fraud risks (Author, 2024)..

   1.3 Description of the Fraud Dataset

   The fraud dataset used in this study is a comprehensive collection of 1,852,394 transactions, which was obtained from the Kaggle website. The data set comprises of 23 variables representing various attributes of financial transactions including both legitimate and fraudulent transactions. The data set also captures transaction amount, customer demographics, geographic location, transaction time, and customer details like, date of birth, occupation, and gender. Table 1 below highlights the important variable that will aid the analysis and detection of fraud.

2. Analytics Design

The analytics design employed in this study follows a structured approach, incorporating various data mining techniques and statistical methodologies. The process begins with data preprocessing, which involves cleaning and transforming the raw data to ensure its quality and suitability for analysis. Subsequently, exploratory data analysis (EDA) is conducted to gain a deeper understanding of the dataset’s characteristics and identify potential relationships between variables. To illustrate the analytics design process, a Unified Modelling Language (UML) activity diagram is presented below:

3. Data Analytics

Tableau Desktop and SAS academy were the analytic used to analyse and carry out the prediction of fraud in this section.

3.1 Data Pre-Processing

Before proceeding with the analysis, several data pre-processing steps were performed to ensure the quality and consistency of the data.

3.1.1 Data Cleaning

The dataset was thoroughly examined for missing values, duplicates, and outliers. Missing values were imputed using appropriate techniques and removed when they constituted a significant portion of the data. Duplicate transactions were identified and removed to avoid redundancy and skewed results.

3.1.2 Data Transformation

Categorical variables were encoded using techniques such as one-hot encoding or label encoding to make them suitable for analysis. The encoded numerical variable is_fraud was replaced by a new binary variable that is for the unique integer value.

3.1.3 Data Subsetting

The data set is about 1.8 million observations and this quite large and will cause performance issues when running the analysis. To resolve this issue, a subset of the data was taken. This was done by looking at the top ten highest transaction by state and the top ten highest fraudulent transaction. New York was the state with the highest fraudulent transaction and had the second highest transaction. Our focus moving forward will be focused on the state of New York.

3.2 Exploratory Data Analysis

To gain insights into the characteristics of the credit card fraud dataset and identify potential patterns and relationships an exploratory data analysis (EDA) techniques were employed.
As shown in the previous fig
• Non-Fraudulent Transactions: Texas (TX) has the highest number of non-fraudulent transactions (134,677) and total transaction amount ($9,239,298), followed by New York (NY) and California (CA). • Fraudulent Transactions: New York (NY) leads in fraudulent transactions both in count (730) and total amount ($395,295), followed by Texas (TX) and Pennsylvania (PA).

3.2.1 Distribution Analysis

EDA was carried out to gain a broad understanding of the dataset’s features and attributes and uncover potential relationships between variables. Summary statistics, such as mean, median, and standard deviation, were calculated for numerical variables, while frequency distributions were examined for categorical variables.

Fig 3.3 Summary Statistics of Gender with regards to transaction population and age.

Fig 3.4 Average spend and number of transactions by gender.

The above table shows the average amount spent by customer gender. On average Male customer spend more money per transaction when compared to female customer and the overall average. However, female customers carry out more transactions and in essence spend more money overall.
To understand the age distribution of the customers using credit card a box plot is shown in the below figure.

In the above figure, it can be observed that on average the male customers are older than the female customer. With the female customer on average being in their forties and the male customers are in their fifties. Another trend that can be observed from the figure is that fraudulent transactions tend to happen in both genders. This can be alluded to the fact that older individuals are more susceptible to falling for scam as they are not technologically as savvy as their younger counterparts. This insight os quite significant as stakeholders in the customer experience team can use this detial and reach out to older customers especially, giving them training on how to spot fraudulent transactions there by curbing the risk of them falling victim to scam.

The above figure shows the frequency of legtimate transaction and the week day in which they occur the most. Monday and Sunday are the days where most transaction takes place and this is regardless of the gender making the transaction. This insight will enable stakeholders allocate resources to begining of the week and ensure there are no failures.

In contrast to fig 3.6 the above figure shows that fraudulent transactions for male customers exceeds that of female customers from Wednesday to Saturday. However, fraudulent transaction mostly take place from Sunday to Tuesday for female customer. The spike days for fraudulent transactions are mostly Sundays and affect female customers the most. This insight can be utilized to educate customers with what to look out for with regards to fraudulent activities on their credit card.

Fig 3.8 gives us insights into customer spending habits. We can see from the table that female customer out spends the male customers in every category except Travel. The highest category where customers spend their money is in Grocery point of sale i.e. where they physically swipe their card at the counter. The least amount spent in a category was also in grocery, but the payment method is online. It will make sense to market travel insurance to male customer and market coupons to female customers who physically shop for Grocery items

Fig 3.9 is a bar graph that splits the total transaction amount into the different category which customers carried out transaction. It can be seen that the bulk of the transaction was done in the grocery category. While shopping online had the most fraudulent transactions followed by miscellaneous shopping online. This is a good insight as stake holders can flag easily a transaction to be fraudulent when it falls into these categories and a warning message alerting the customer can be sent.

To understand the spending habit of customers the table in fig 3.10 was created, it shows that personal care and online grocery shopping are the two least categories where customers spend their money. On average, the least amount spend by a customer is on personal care but the most amount spent on average falls within the travel and physical grocery category. When compared to the median spent by customers the story slightly changes, miscellaneous and shopping but in physical stores and online shopping categories were the least amount were customers spend their money.

3.2.2 Time Series Analysis

To understand the difference in characteristics of fraudulent transactions and non-fraudulent transactions, a time series analysis of both kinds of transactions was carried out. This will create a clear differentiation between what is considered to be fraudulent transactions and what is not.

The figure above shows fraudulent transactions plotted by the hour of the day in which the transaction occurred. Instantly a theme can be seen, the scammers that carry out fraudulent transactions do so in the late hours of the day into the early hours of the day. This kind of information already gives a good red flag of when transactions can be halted if suspected of fraudulent activities.

Fig 3.12 shows the hours in which customers carried out transactions by both male and female. There is an even spread of the data, with peak periods starting at mid-day and least transactions happening at midnight. When fig 3.12 is compared to fig 3.11, it is clear how fraudulent transactions can be profiled and as such detected using the hour of the day as a feature in a model that will predict fraudulent transactions.

3.2.3 Correlation Analysis

To effectively predict fraud, a correlation analysis was carried out using the amount spent by a customer and the likelihood of fraud to be committed. To achieve this, the amount being spent by customers were broken down into bins of eight. Each been representing a fraction of the total amount spent. The range of the bins are $20, $40, $100, $200, $500, $1,000, and above $1,000. The table shows that as the amount of money spent increases so does the likelihood of the transaction to be fraudulent. Fig 3.13 Table showing increase in amount spent and the likelihood

Our earlier analysis showed that the average amount spent on any transaction by a customer, either female or male was $71, however when you look at fig 3.14, which is a characteristic of fraudulent transactions, the observation is that fraudulent transactions tend to be three times or higher the average spend of a customer. This makes sense as scammers would want to cash in big in order to get away with as much more as they can.

3.2.3 Customer Age Analysis

At the beginning of our analysis, we did a distribution of age within the customer base using the credit card. This showed that older customers tend to get defrauded. The below table is a comprehensive analysis of age of customers and the likelihood of getting defrauded. Just like we did with the correlation analysis, we will group the customer base into age range. Teenagers, customers below the age of 20 years old, Young Adults, customers between the age of 20 – 35 years old, middle age between 35 and 60 years old, and then lastly Elderly, customers older than 60 years. This grouping will also be further grouped into gender, so we can have a clear view of how the age structure is between genders as well. The below figure shows in clear details this analysis.

Imploring a bar graph to assist in buttressing the insight of our analysis, we can see that middled aged male customer are the most defrauded within our customer base followed by elderly women and young women. Teenage male customers also tend to fall victim as well.

Conclusion

This study aimed to analyse a credit card fraud dataset to gain insights into patterns and characteristics that can aid in the detection and prevention of fraudulent activities. Through exploratory data analysis using SAS and Tableau, several key findings were uncovered. The analysis revealed that certain factors, such as transaction amount, customer age, gender, and time of day, can be indicators of potential fraud. Fraudulent transactions were found to be more prevalent among older customers, suggesting a need for targeted education and awareness campaigns. Additionally, fraudulent activities were more likely to occur during late night and early-morning hours, providing a temporal pattern that can be leveraged for fraud detection models. The critical comparison between SAS and Tableau highlighted the strengths and weaknesses of each tool concerning different aspects of fraud analytics. While SAS excelled in data manipulation, statistical analysis, and model building, Tableau demonstrated its prowess in interactive data visualization and user-friendly exploration. The insights gained from this study can be valuable for financial institutions and other organizations in developing robust fraud prevention strategies, optimizing resource allocation, and enhancing customer education and support.

Appendix 1 – Coding SAS Code

Data Pre-processing

Appendix 2 Tableau Visualisation

Data Pre-Processing

Data Mining