Predicting home network problems using diverse data

Providing uninterrupted high quality service is very important for service providers to avoid customer churn and to minimize the cost of customer care. Predicting service disruption and degradation, followed by proactive corrective action, helps service providers mitigate issues before they are noticed by customers. In this paper, we present a framework and a set of algorithms for the prediction of home network problems using a diverse set of data sources. More specifically, we discuss data collection, pre-processing and model building steps as applied to various data sets arriving from home network devices such as network interface devices, home routers, and customer care systems. We also present the results of a performance evaluation study where we applied our framework to an anonymized telecom data set. For this data set, our techniques were able to predict 75% of the “cannot connect to internet” problem, which was the top call driver to customer care.