Optimization of Delay Tolerant Routing using Machine Learning Techniques

This paper presents the machine learning approach to improvise the routing decisions in DTN routing. Routing strategies in DTN is basically to choose a suitable nodes to transfer message copies for the successful transfer of messages to the destination node. The routing decision, whether to transfer the data to the node encountered or not is considered similar to multi-label classification problems. We have used five multi-label classification techniques for finding the optimum technique for this task on Zebranet UTM-1 data for PRoPHET Routing and Epidemic Routing. The techniques used are Ensemble chain classifiers (ECC), CC(Chain classifiers), BR(Binary relevance), Label Power-set problem transformation and OneVsRest. We have used 5 classifiers as the base learners like XGBoost, k-NN, D-Tree, Naive Bayes, and Random Forest, . The library used for parameter optimization of the classifiers are hyperopt library and GridSearchCV. Ensemble chain technique with XGBoost classifier on PRoPHET routing data outperformed all the techniques for the PRoPHET and epidemic routing with an accuracy score of 96.1%, Jaccard score of 92.08% and f-1score score 94.47%.