Efficacy of Tree-Based Classification Models for Intrusion Detection in IoT Networks

The proliferation of Internet of Things (IoT) has made it possible to envisage distributed applications which can be better managed. The edge devices of an IoT network can capture real-time data and transmit over a network to strategic points where it can be quickly analyzed for timely response. Applications such as industrial control, smart cities, smart firming, supply-chain monitoring can largely benefit from such IoT-based networks. But such applications are vulnerable to network attacks as mischief mongers may try to disrupt the activities by sniffing the data while on transit or illegally access the servers to manipulate the data. Therefore, protecting IoT networks from intruders is a critical concern in order to avoid disruptions in the execution of activities. One of the approaches to achieve this objective is to train machine learning models with the available data and test their performance under different scenarios. As the models fine-tune themselves over a long period of time by learning from varieties of data, they begin to perform better as time progresses. In this work, 24 machine learning models were developed by combining each of the six tree-based classifiers, viz., SPAARC, Random tree, BF tree, Random Forest, Simple Cart and J48 with four feature selection algorithms, viz., Relief F, Symmetric uncertainty, Gain Ratio, Chi-squared. The relative performance of the models was evaluated using fifteen measurement criteria, viz. precision, F value, FPR, precision, recall, specificity, MCC, KC, NPV, FNR, GM, YI, RAE, MAE and RMSE. Extensive experiments showed that the relief-F + random tree model demonstrated the best performance with respect to most of the evaluation metrics with an accuracy of 0.9995, F value of 0.9997, precision of 0.9997, and lowest false positive rate of 0.0009. On the other hand, the BF tree model showed the worst performance among other models irrespective of the choice of feature selection algorithms.