際際滷

際際滷Share a Scribd company logo

Less sketchy

S
Shih-Ho Cheng

The document describes LessSketchy, a machine learning tool that analyzes online posts to warn users about scams. It trains a balanced random forest classification model using a large dataset of posts scraped from the web. The model is trained to identify scams versus legitimate deals by balancing the dataset and bootstrapping trees from balanced samples to address the class imbalance between scams and real posts. The tool aims to help users avoid scams when searching for deals online.

1 of 14
Download to read offline
LessSketchy by Shih-Ho Cheng
Many posts seem to be great deals ...
Many posts seem to be great deals ... Cheap! Great Location No Credit Check
... but many are really scams!
... but many are really scams! How can I prevent this?
LessSketchy uses machine learning to warn you about scams
Where to get training set?
Where to get training set? 1. Web-Scrape a large sample
Where to get training set? 1. Web-Scrape a large sample 2. Wait...
Where to get training set? 1. Web-Scrape a large sample 2. Wait... 3. Re-visit each post and check for removal
Balanced Random Forest Use it to train classification of scam vs. legit
Balanced Random Forest Use it to train classification of scam vs. legit Deals with an imbalance sample (1 : 99)
Balanced Random Forest Use it to train classification of scam vs. legit Deals with an imbalance sample (1 : 99) Each tree is trained by bootstrapping a balanced training sample
Balanced Random Forest Use it to train classification of scam vs. legit Deals with an imbalance sample (1 : 99) Each tree is trained by bootstrapping a balanced training sample Aggregate the classification of each tree

More Related Content

Less sketchy