Finding Hidden Links among Variables in a Large-Scale 4G Mobile Traffic Network Dataset Using Machine Learning

Using a mobile dataset from Orange Senegal small cell 4G network, we study the effect of variables such as data traffic at the downlink level, data traffic at the uplink level, total data traffic, maximum number of active users, signaling protocol, uplink user rate, downlink user rate, physical resource block rate for downlink, block rate physical resources for uplink, load data logging, channel quality indicator, downlink radio delay average, over the perceived rate at the downlink. We are looking to find the variable that most affects the perceived flow. We do this by using machine learning to find the variable that closely explains the variation in perceived flow and helps predict flow with greater accuracy. We observe that correlation analysis is unable to find a hidden relationship between throughput and other variables. With models such as linear regression, decision tree, random forest and multi-layered perceptron, the channel quality indicator (CQI_Avg) turns out to be the variable that more closely explains the variation in perceived flow and more accurately contributes to the prediction compared to others variables.