Call Drop Prediction Using Nonlinear Autoregressive with Exogenous Input(NARX) Model

Abstract

In the global system for mobile communication (GSM), call drop is one of the key parameters for performance indicator (KPI) as it affects customer satisfaction. This research presents a non-linear autoregressive exogenous (NARX) Neural Network model for predicting a call drop due to handover failure. The three handover failure parameters: handover failure due to no response from originating and destination side (HoF1), handover failure due to master switching centres (MSC) route selection failure (HoF2) and handover failure due to call release during handover (HoF3) were used to train and test the NARX Neural Network. Call drop target variable was also used. Four different input sizes (60, 120, 180 and 240) were used each to train the network; this was done to determine the appropriate size of input for training the network. The trained network with 120 inputs showed better performance in terms of means squared error (MSE) and a high coefficient of regression (R), hence it was adopted for predicting call drops for 20 hours ahead of the 120 hours used for training the network. The model was implemented using the time series Neural Network prediction in the MATLAB. The result shows a coefficient of regression value of 0.88653, prediction accuracy of approximately 89% with an MSE of 4.06697. The result will help telecommunication companies in improving the quality of service with the knowledge of call drops that are likely to occur in the future for a particular area, thereby improving customer satisfaction.