A Predictive Handover Approach in LTE Networks with Measurements and Decision Tree Algorithms (Case Study City of Quito)

Abstract

This work analyzes the handover process in LTE (Long Term Evolution) networks in two scenarios, detecting the features and behavior of RF (Radio Frequency) parameters by means of predictive models. The two scenarios collect measurement data of RF parameters, the first one in an urban area of Quito city, in order to analyze the behavior and establish the features of an area with handovers. The second scenario seeks to obtain a predictive model of the failed handover zones; in this case, we make tours in a rural area of Quito city and analyze the failure of handovers with failures in VoIP calls, since in this service the loss of connection in LTE is more noticeable. For data collection, monitoring tools installed in cell phones are used and for data analysis and obtaining the Machine Learning model, R and RStudio are used. The collected data were cleaned and transformed to obtain unique DataSet for each scenario, then divided into a training and test set. The training set was processed using the decision tree technique, which allowed obtaining a graphical model of the behavior of the RF parameters that generate handovers or failed handovers, according to the scenario. Finally, the models were evaluated with the test set by defining confusion matrices and calculating the respective accuracy, with 96.34% in scenario 1 and 95.5% in scenario 2.