Classifying wireless IOT ICU traffic with machine learning models

Abstract

In this work, we classified the wireless internet of things (IoT) traffic of the IoT Health intensive care unit (IHI) dataset which belongs to three general classes: patient monitoring, environment monitoring, and network attack. We trained and tested 7 machine learning (ML) models with Orange 3 including kNN, Decision Tree (tree), SVM, Random Forest (RF), Neural Network (NN), Gradient Boosting (GB), and AdaBoost (AB). With the original dataset, 5 ML models performed perfect classification. After pruning the dataset columns by keeping the features with the highest correlations with the label in the dataset, good classifications were obtained with only 4 TCP/IP features by the Gradient Boosting, kNN, and RF models with MSEs in the range 0.008-0.011, and R²s in the range 0.978-0.984. With only 6 MQTT features, Gradient Boosting, RF, Tree, and NN were the top classifiers with MSEs in the range 0.073-0.074, and R²s in the range 0.856-0.859. This work demonstrates the effectiveness of guiding the feature pruning process by the values of the correlation coefficients in order to minimize the long training times of ML models while achieving good accuracies.