๐Ÿ”’Machine LearningTrending

Detecting Fraudulent URLs with Machine Learning

The engineering story behind SafeCheck โ€” from feature engineering to model selection, deployment on Streamlit, and lessons from building production ML pipelines.

Mohamed Mydeen

Mohamed Mydeen

Full Stack & AI Engineer

March 10, 2025
11 min read
#ML#Python#Security#Streamlit#Feature Engineering

The Problem Space

Seasonal fraud websites are a specific and underserved threat vector. They appear during holiday seasons, mimic legitimate e-commerce platforms, and disappear before traditional blocklists catch them. SafeCheck was built specifically to catch these.

Feature Engineering

The most important insight: domain age and SSL certificate issuance date together are the strongest predictors of fraudulent intent.

python

Model Selection

After benchmarking 7 algorithms, XGBoost with calibrated probabilities won on both accuracy (94.2%) and interpretability โ€” crucial for an explainable AI system.

python
0 Claps

Limit 50 claps per article

Mohamed Mydeen

Mohamed Mydeen Shahabudeen

Full Stack & AI Engineer

Building AI-powered products and writing about the engineering behind them. Creator of Mydeen AI, Feast At Night, and SafeCheck.