Using Neural Networks and Machine Learning to explain Impact of Predictors in a Classification Scenario

Abstract

This paper as an attempt in predicting the winning probability of a deal using two different methods multinomial logistics model and neural networks (NN) model. The objective of the study is to find out the relevant predictors with their important score for each of the outcomes of the dependent variable. Using Machine learning method and the traditional statistical modeling method we compared the results and found the machine learning method more appropriate. With the advances in emerging fields, it is plausible that newer techniques like NN and Machine Learning (ML) may finally be able to not just provide predictions that are more robust than conventional methods, but also explain how each predictor variable impacts the prediction. The data were analyzed using open source software tool R (R-Studio). One of the key findings of the study was not only to arrive at a better overall model accuracy, but also it looked at the outcome value taking into consideration the business interest. This exercise is an attempt to arrive at a methodology using such a combination to not only predict which deals being pursued by the sales team of a healthcare intermediary are likely to result in a desired outcome (Win), but also show how each factor will impact the outcome, and to what extent. This would not only enable the users to determine which deals to prioritize but also determine what specific actions they need to take to achieve this. This could open new doors for B2B sales analytics in general and the Healthcare sector in particular.