I will build a fraud detection machine learning model in python

Pakistan

I speak Urdu, English

Full Stack Web, DevOps and AI Automation, The Burujan

The Burujan is an engineering studio run by Huzaifa, a full stack developer and NOC Engineer with a BS in Computer Science. We build fast Next.js websites, custom dashboards, CRMs and APIs, and Flutte...
About this Gig

Fraud losses hurt, but so do false alarms that block good customers. I build a fraud detection machine learning model in Python that catches more fraud and flags fewer real customers.


I am Huzaifa from The Burujan, a NOC engineer who builds data science projects. My credit card fraud detection project uses 590,540 real card transactions where only 3.5% are fraud. On a later test set the model never saw, it reached 0.91 ROC AUC, caught 71% of fraud dollars, and cut false alarms by 47% compared with a standard model at the same fraud dollar capture.


What you get: data cleaning, behaviour features such as spend history and velocity, class imbalance handling, tuned LightGBM and XGBoost models, a decision threshold based on your cost of fraud versus a false alarm, SHAP explanations so analysts see why each transaction was flagged, and a report with a Jupyter notebook. Premium adds a scoring API, monitoring plan and retraining guide.


I split data by time, not randomly, so results are honest. No model can promise zero fraud, and results depend on your data. Message me before ordering with a short description of your data and I will tell you if it fits.

Expertise:

Classification

•

Anomaly detection

•

Predictive analysis

Programming language:

Python

•

SQL

Frameworks:

Scikit-learn

Tools:

Jupyter Notebook

•

Colab

My Portfolio