Lgbm-Customer Returns Prediction

This project was developed by Doğu Can ELÇİ for Kaggle-Trendyol Challenge as last submit in detail.

Competition details:

The data for this challenge consists of user and product metadata, and user transactions. The user transaction data is divided into two parts: training period (from 2021-05-01 to 2021-07-31) and test period (from 2021-08-01 to 2021-08-07). Within the training period, the information whether a bought product is returned or not is provided. On the contrary, this information is missing for the test period.

The main task of the project is to predict whether the user will return any of the products he/she has bought in a single session having the same productcontentid.

File descriptions:

- test1_model: feature selection,model training and evaluation
- test2_products: product information processing
- test2_pure_data: merge process of all prepared and edited dataframes from other notebooks
- test2_reviews: processing of user comments and rankings for products
- test2_user: user demographic information processing