RECOMMENDATION ENGINE FOR B2B CUSTOMERS IN TELECOM BY CUSTOMIZING KNN ALGORITHM

Abstract

K-Nearest Neighbours (KNN) is one of the most popular algorithms for classification; however traditional KNN algorithm has two limitations: 1. There is no weight difference between closest training examples 2. Revenue Prediction is not feasible In this paper, we propose a KNN type method for classification that is focussed at overcoming above shortcomings. Our method constructs a cross-sell penetration model using Revenue, Usage, and Firm graphics data for targeting telecom Enterprise Customers. Value of K is varied for different data, and is optimally chosen based on classification accuracy. After Propensity of an account is determined from traditional algorithm, weights are assigned to nearest neighbours and Revenue is determined.