Improving Customer Experience for an Internet Service Provider: A Neural Networks Approach

Abstract

Today one of the challenges of companies is to decrease call center costs while improving the customer experience. In this study, we make prediction and proactively take action in order to solve customer problems before they reach the customer call center. We use machine learning techniques and train models with a dataset of an internet service provider’s several different systems. We first use supervised techniques to classify the customers having slow internet connection problems and normal internet connection. We apply two classification approaches, multi perceptron neural networks and radial basis neural networks. Then, we cluster the same dataset using unsupervised techniques, namely Kohonnen’s neural networks and Adaptive Resonance Theory neural networks. We evaluate the classification and clustering results using measures such as recall, accuracy and Davies-Bouldin index, respectively.