Skip to content
  • ·ContentsHome. The title page, every section at a glance.
  • IWritingEssays and longer notes. Slow to write, slower to finish.
  • IIWorkThings I've built. Small, finished, and used by at least one person (me).
  • IIIInspirationThings I want to look at again.
  • IVLinksEverything I read and want to keep, on one page.
  • VQuotesSentences I pressed between pages.
  • VIPlaygroundExperiments that move.
  • VIINowWhat I'm doing this season.
  • ·ColophonHow this book is made: type, plants, sound.
↑↓ move · enter open · esc closetype to search every entry
Collected
01-28-2024
Reading
1 min
Family
#machine-learning · #python

Loan risk analysis

View source on GitHub

This project analyses loan applications and builds a model to predict whether an applicant will repay in full or default. The public repository records a 93.07% evaluation accuracy and includes the analysis notebook used to build the LightGBM model.

The broader workflow covers data preparation, exploratory analysis, feature engineering, model training, and evaluation across 577,000 loan records.

The useful lesson is not the headline score in isolation; it is the end-to-end discipline of moving from raw borrower and loan attributes to a reproducible evaluation workflow.

Related: Applied ML