Root cause detection of call drops using feedforward neural network

摘要

Call drop rate in GSM (Global System for Mobile Communication) network is an important key performance indicator (KPI) that directly affects customer satisfaction. The delay in identification of exact call drop reason because of multiple reasons involved in it would results in poor customer satisfaction. The TCH (traffic channel) call drops due to three different hardware causes are collected from live GSM network for 10 days and are represented in time domain. Time domain features such as mean, maximum, standard deviation etc. are extracted from each type of call drop signal which is used to train the feedfoward neural network. FF neural network is made as decision making classifier, feature vector is inputted and root cause detection information is outputted.