Picking a Threshold Is a Business Decision (Day 6)
The chart doesn't decide for you. I had to.
Oct 4, 20264 min read
Command Palette
Search for a command to run...
Latest articles
XGBoost Caught More Fraud - But There's a Catch (Day 5)
My recall jumped from 59% to 81%. My precision dropped from 79% to 57%.
Oct 4, 20264 min read23I Finally Understand AUC, Confusion Matrix, and Classification Report (Day 4)
Day 3 was splitting. Day 4 was my first actual model : a baseline logistic regression. The model trained fine. The AUC looked great. But the confusion matrix and classification report are what finally
Sep 29, 20264 min read20Splitting Fraud Data Without Breaking the Ratio (Day 3)
Why stratify=y is the most important argument in my train/test split.
Sep 29, 20263 min read13What I Learned From My First Look at Fraud Data (Day 2)
Day 1 was setup. Day 2 was the first time I actually looked at the data and the numbers immediately broke my assumptions. Here's what the credit card fraud dataset told me. The problem it solves I had
Sep 27, 20263 min read17How I Pulled a Kaggle Dataset Into Colab (Day 1 of My Fraud Project)
I'm building a credit card fraud detection project from scratch. But before I could touch any data, I had a smaller problem: how do I actually get a Kaggle dataset into my notebook? My old method was
Sep 27, 20263 min read17