Detecting Attacks on IoT Devices in Ambient Intelligent (AMI) Home Network

Abstract:

This study developed an ambient intelligent home framework with a view to detect attack against IoT devices in the home. A framework was designed using an IDS Server, remote users, home users, attackers and IoT devices in ambient intelligent home. Algorithm was developed using stacking ensemble, which include naïve bayes, support vector machine, ada boost and max voting. Botnet dataset that captured different types of IoT network traffic in an ambient intelligent home was used. The stacking ensemble algorithm was evaluated using accuracy, precision, KAPPA, FPR, FNR, FDR F1 score, NPV, MCC, recall and specificity and benchmarked with the individual algorithms. The simulation was performed using ANACONDA Navigator and JUPYTER Lab. The result shows that stacking ensemble could achieve as high as 96.67 percent accuracy in detecting an attack against IoT devices and also 91.67% precision and recall respectively while the algorithms; Naïve bayes, SVM, ada boost, random forest, max voting had accuracy of 73.13 %, 81.86%, 62.71%, 91.61%, 84.92% respectively. A prototype of the deployed ensemble stacking classifier that predicts the category of a network packet in real time was developed using Python programming language. Various Python libraries such as pandas, NumPy, streamlit, joblib, and scikit learn were employed. The program runs 20 times, each with 500 different sets of Botnet dataset. The simulation result recorded average accuracy of 95.57%. This revealed that the framework will effectively protect the ambient intelligent home against intruders and attackers.

Keywords: IDS Server, Ambient Intelligent Home (AmI), IoT devices, stacking ensemble, remote users, prototype