A Model for Clustering and Predicting Customer Behavior in a Structured Data Set

Abstract

The main goal of the work is to study the process of leads coring. In this paper, two ways of lead generation are explained – manual and using predictive models. This condones, per investigation, is the most efficient way to score a lead.The analytical review starts with an overview of existing lead scoring methods, describing each and comparing the pros and cons. There is an analytical comparison of the products that already exist on the market, such as HubSpot, Infer, PipeCandy and Maroon.ai. As a conclusion from this comparison, the effect of this study has a benefit in that it is a low-weight plugin with a significantly lower price than the market median. To provide a list of methods and algorithms used and describe why each was selected for a specific goal. Four aggregations were made to choose the best-fitting model, resulting in the LightGBM selection. There is a comprehensive description of the LighGBM features and math background, along with an explanation of GOSS. Also, after the model was selected, the set data manipulations were done, such as aggregation, class weight balance, tuning and exhaustive analysis, and correction of disbalance. In part to the results, there is an extensive analysis of the model performance after studying the data set and usage in a real-case CRM – Salesforce. As results showed – the created plugin can be easily integrated into any CRM solution using the native marketplaces or package-delivery systems.