What I Learned Analyzing 20,000 Airbnb ListingsThis was my first market analysis project. I had done fraud detection before. That was a classification problem: predict a label. This was different. There was no label. No target. Just a question and12SOct 15 · 5 min readbeginnerslog.hashnode.dev
My Final Fraud Model (Day 7)Day 1 was setup. Day 2 through 6 was the journey. Day 7 was the test set. I had not touched it since Day 3. That was the point. The test set is the sealed envelope. The one number you only open at the1Oct 8 · 5 min readbeginnerslog.hashnode.dev
Picking a Threshold Is a Business Decision (Day 6)Day 5 gave me a model that catches 81% of fraud but at a cost: 45 false alarms. Day 6 was about choosing a threshold, and it taught me that this isn't a math decision. It's a business one. The problemOct 4 · 4 min readbeginnerslog.hashnode.dev
XGBoost Caught More Fraud - But There's a Catch (Day 5)Day 4 was a baseline logistic regression. It got 0.9571 AUC but missed 40% of frauds. Day 5 was about fixing that. I switched to XGBoost and used a trick called scale_pos_weight to handle the class imOct 4 · 4 min readbeginnerslog.hashnode.dev
I 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 finallySep 29 · 4 min readbeginnerslog.hashnode.dev
