Forecasting and Analysis of Online Orders Rejections Using Data Mining Algorithms and Penalty Function

The direction of products and food delivery from online stores and similar transportation services have become especially relevant in recent years in connection with the rapid informatization of society, and especially because of quarantine restrictions due to the worldwide pandemic. However, along...

Full description

Saved in:

Bibliographic Details
Published in:	2021 IEEE 3rd International Conference on Advanced Trends in Information Theory (ATIT) pp. 286 - 289
Main Authors:	Suprun, Oleh, Klimenkova, Nina, Melnyk, Maksym, Lavriy, Sonya
Format:	Conference Proceeding
Language:	English
Published:	IEEE 15-12-2021
Subjects:	Companies data mining machine learning Mobile handsets neural networks Pandemics penalty function Predictive models Profitability Training Transportation
Online Access:	Get full text
Tags:	Add Tag No Tags, Be the first to tag this record!

Description
Summary:	The direction of products and food delivery from online stores and similar transportation services have become especially relevant in recent years in connection with the rapid informatization of society, and especially because of quarantine restrictions due to the worldwide pandemic. However, along with increasing demand, online stores and individual couriers often incur significant financial losses due to customer rejection of orders, disruption in preparation, or other resignation issues that are independent of the company. Therefore, the article presents a way to partially reduce financial and time losses by analyzing preliminary order statistics, identifying potential dependencies and, consequently, predicting future customer rejection based on order details. Given the large number of statistics, it is advisable to use the tools and paradigms of data mining, two different approaches are considered and compared. At the same time there is a problem of false positive conclusion of the model, due to which the system ignores the profitable orders. To reduce such losses, the use of the method of penalty functions is proposed, which allows to assess the model risks and possible benefits. The described approaches were tested on a large real sample, and the corresponding results are presented, which demonstrate the effectiveness of the approach.
DOI:	10.1109/ATIT54053.2021.9678868