Decision Tree based Classification of Profiled Mobile Device Resource Status Information for Data Offloading in Private Network

Abstract

This paper validates the effectiveness of decision tree classification models by performing the data analysis of the large data set collected in the real time implementation of an intelligent composite offload decision (ICODA) framework with on-premise mobile device cloud. We have collected around 40000 data records of real time device resource status information with 7 inputs and 1 output attribute each at different time conditions. These 40000 data records are then cleaned and normalized to scale down in practical range to about 4608 training samples. The machine learning classification technique is applied on different train-test-split ratio using ID3, CART and random forest classifiers (RFC) with proper randomized and grid search cross validations. The resulting mean accuracy percentage is observed at 92.91 with ID3, 99.22 with CART and 99.44 with RFC evaluating all the possible combinations in the data set. The experimental results show that the random forest classifier outperforms the other methods in data offload framework.