AHSWN Home · Issue Contents · Forthcoming Papers

Attack Detection on Internet of Things via Deep Ensemble Classifier Model with Proposed Feature Set
Rekha H and Siddappa M

Due to the large-scale application & difficulties, IoT is now a prominent subject of research. Initially, detecting the attacks and the malicious traffic is a major problem because of traffic network size. However, with the continuous expansion of its size & applicability, security is a major problem. Individual installment of security measures for each IoT device during new threats arising is indeed a time-consuming task. Yet, existing security measures only address the restricted attacks, as evaluations were limited to outdated datasets. This paper offers a novel attack detection model for IoT that will be implemented in the given stages: “(1) preprocessing, (2) feature extraction, & (3) classification”. The challenge of minority class is initially addressed by defining a new logic during the pre – processing stage. Here, a feature extraction stage is applied on the image input (preprocessed data) to extract statistical (Mean, Median, Mode, SD), higher cognitive statistical features (Angular moment, Skewness, Homogeneity, Percentile, Kurtosis), enhanced Correntropy, and better correlations texture methods. The categorization processes are then applied to these characteristics using deep ensemble classifiers including Convolutional Neural Network 1 (CNN 1), Convolutional Neural Network 2(CNN 2), and QNN (Quantum Neural Network). The results of the classifiers (CNN1, CNN2, and QNN) is progressed with enhanced score level integration is used to calculate the outcomes. The proposed ECISF scheme for learning percentage 90 has, respectively, greater specificity of 58.33%, 27.08%, 33.33%, 14.58%, 72.91%, 35.41%, 89.58%, and 68.78% to current schemes like ESFCM, IoT-IDCS-CNN, Bi-GRU,LSTM,SVM, DBN, RNN, and RF.

Keywords: Internet of Things; Attack Detection; Improved Score Level Fusion; Convolutional Neural Network; Quantum Neural Network.