A leading bank flagged several customer accounts as potential money mules and enlisted a data scientist to uncover patterns behind this behavior. Using synthetic data from Experian — including Account Data, Account Holder Data, and Mule Flag indicators — the dataset underwent cleaning (handling duplicates, imputing missing values with mean/mode) and feature engineering (binning ages). After merging by ID, the data was prepared for exploratory analysis.
To identify and summarize key predictors for accounts more likely to be involved mule activities.
-
Based on gender, The split is nearly even — 29 female against 30 male mule accounts.
-
There are no mule accounts within the youth (18-24 years).
-
Mule accounts were most common among ages 55-64.
-
Surprisingly, high employment rate was seen among individuals aged 65-74 despite the UK retirement age of 66 years - raises possible red flag.
-
Married account users were flagged more than the singles.
-
Higher mule risk is found among employed, and retired individuals, as well as male students.
-
Older people reported longer hours on social media which could relate to manuiplation or vulnerability.
-
A mild upward trend exists between withdrawals and trnasfers - more withdrawal frequency correlates to more tranfers.
V1: Mule Accounts by Age, Gender, Employment
V2: Marital Status of Mule Accounts
V3: Employment Status by Age Group
V4: Social Media Usage Hours by Age Group
V5: Transaction Behaviour by Mule Status
Based on the patterns above, the strongest predictors include:
- Employment status - particularly retirees, married individuals, and male students.
- Marital status - married users show higher involvement.
- Age group - especially those from 45-74.
- Social media usgae - long hours observed in higher risk users.
This project, built using Python (Pandas, Seaborn, Matplotlib), illustrated early-stage indicators of mule account behavior. These insights can inform the bank’s security protocols and fraud detection systems. Furthermore, next steps could involve training and deploying a machine learning model to automate mule account detectio with higher precision.