Intelligent IoT Network Malware Classification using Realtime Heterogenous Data
Δεν υπάρχει διαθέσιμη μικρογραφία
Ημερομηνία
2023-06-07
Συγγραφείς
Τίτλος Εφημερίδας
Περιοδικό ISSN
Τίτλος τόμου
Εκδότης
Δικαιώματα
Default License
Άδειες
Παραπομπή
Παραπομπή
Περίληψη
Περίληψη
Due to its wide range of applications, the Internet of Things (IoT) technology is evolving
rapidly. One can witness IoT systems in smart cities, smart homes, smart healthcare, smart
industry, and smart agriculture. IoT systems usually use low-powered and low-memory
devices to sense the data from the environment and transmit it to the destination through
wired or wireless communication channels. Although IoT technology is gaining massive
attention in every sector of life, the security of these devices is one of the biggest challenges. Due to resource constraints, these devices are often vulnerable to malicious actors.
In this work, a machine learning-based intelligent classification of the IoT network attacks using real-time heterogenous data is carried out. Two IoT network malware datasets
(Ton-IoT & IoT-23) that include the real-time IoT Botnet attacks are used for the experiments. The data is pre-processed before performing the experimentation. In addition, a
information gain based feature selection method is also applied to select the most important features in the dataset. Several classification methods include Logistic Regression
(LR), Decision Tree (DT), Random Forest (RF), K-Nearest Neighbors (KNN), Naïve
Bayes (NB), and eXtreme Gradient Boosting (XGB) are implemented. These models
were evaluated using classification metrics; accuracy, precision, recall, and f1-score. It is
concluded that the Naïve Bayes and Logistic Regression are not the best methods to perform classification on these datasets. On the other hand, DT, RF, KNN, and XGB provided an accuracy of 99% for binary labels and 98% for multiclass labels for the Ton-IoT
dataset. Using the IoT-23 dataset, these models provided accuracy above 90%. It is found
that LR and NB are not the best choices for classification using either dataset. In addition,
not all the features in these datasets are essential; hence some can be dropped to reduce
the complexity of the model and improve the computational capacity. It is further concluded that heterogeneity in the dataset does not necessarily affect the performance of
classification algorithms.
Περιγραφή
Λέξεις-κλειδιά
IoT malware, Heterogeneity, IDS, Classification, BotNet Attacks