Lenta-Hackathon

Code and files for lenta hackaton September 2020.

All details you can find in a full presentation.

Prepocessing

We analyzed data, clean and split it for better algorithm work by following factors:

1. Checks with big amount of different items (big chance to leave if you haven't found what you need)

2. Type of plate (different customers behavior pattern)

3. A few most popular items and all low popular items (first - outliers, second - not informative)

4. Transactions in which there was a wholesale purchase goods (for example сafe owner buy 100 liters of Coca Cola - outliers)

PageRank

Graph example: