This paper describes a KPI Predictor: A Machine Learning based approach to predict KPIs based on underlying counters. Prediction or forecasting KPIs is important because network level activities bank on it. These are situations when changes at the network or on the mobile handset are planned to be played but the Telecom Service Providers would like to know its exact impact so as to take preventive steps. A few scenarios are:
  • The call volumes in a cell will decrease by 10% owing to the activation of a particular feature on the network side. So, what would be the impact on Call Success KPI?
  • The neighboring cell will increase the coverage (power) by 10%, will there be an improvement in the Call Drop KPI?
  • A power-saving feature will be enabled on the mobile phone OEM chipset making the mobile phone release its resources 10% earlier. Will this result in unused capacity on the cell?
Given the technical challenge in KPI troubleshooting, it follows that predicting/forecasting KPIs is a very uphill and convoluted task though it is a very important and necessary one. With this background, this paper tries to see how Data Science can be used in such a way that SMEs have more qualitative and quantitative insights into KPI trends; and thereby build models.