The Normalized Mean Value Model: A Novel Approach For Propensity Score Prediction And Its Comparative Performance

Abstract

This research paper compares Logistic Regression, Random Forest, and a new proposed algorithm ”Normalized Mean Value Model” to see how well they predict propensity scores for different events based on website activity data. The goal is to find the model that gives the most accurate propensity scores for website product purchases. This could then be used to predict other likely events. The paper explains what propensity scores are and what data is needed to calculate them from website activity and compares how well Logistic Regression and Random Forest models perform against the proposed models, termed ”Normalized Mean Value Model” built using correlation strength of independent variables. This research aims to validate the effectiveness of the proposed model propensity score calculation and ascertain whether the distribution of the calculated propensity scores is solely influenced by specific correlated independent variables. Additionally, this research seeks to uncover potential improvements and insights that could enhance existing prediction models.