An Improved Intrusion Detection System Using Machine Learning with Singular Value Decomposition and Principal Component Analysis

Abstract

Among the most crucial components of a cyber physical system is a network of nodes via which a large number of autonomous, mobile or stationary sensors can communicate with one another. This network is known as a wireless sensor network (WSN). The network's nodes work together to sense the world around them, collect data on the items they detect, process that data, and then communicate it on to the network's owner since WSN has many potentials uses across many disciplines but little available resources, it is often coupled with IOT, making the network accessible to the outside world and susceptible to cyberattacks. Blackhole, grayhole, flooding, and scheduling attacks are some examples of common attacks in WSN that can do significant damage rapidly. Intrusion detection approaches for WSN suffer from issues like a low detection rate, a large calculation overhead, and a high false alarm rate because of the network's redundant and highly correlated data and the constraints imposed by sensor nodes' limited resources which is the research problem. This research proposes a solution, dubbed IDS-ML, that makes use of three different machine learning techniques-stochastic gradient descent (SGD), ridge regression (RR), and gaussian naive bayesto solve the problem of intrusion detection in wireless sensor networks (GNB). In order to reduce the computational burden of the technique, principal component analysis (PCA) and singular value decomposition (SVD) are applied to the original traffic data to lower the feature space dimension. Once network threats have been identified, an IDS-ML model is utilized to categorize them. Based on the experiments with two datasets WSN-DS and UNSW-NB15, the proposed IDS-ML achieves significantly higher accuracy rate of 99% than state-of-the-art detection algorithms for WSN-DS and UNSW-NB15 dataset. As the achieved higher accuracy rate against normal, blackhole, grayhole, flooding, and TDMA attacks are 99%, 100%, 72%, 100%, and 78% respectively.

A Hybrid Cuckoo Search-K-means Model for Enhanced Intrusion Detection in Internet of Things

“…The evaluation of our approach on the dataset utilized the confusion matrix (CM), expressed in Eqs. ( 11)-( 14), where True Positive (TP) and False Negative (TN) counts are balanced against False Positive (FP) and False Negative (FN) counts [35]. Metrics such as precision, accuracy, and the F-score are considered to assess the classification model's performance comprehensively.…”

Stator Imbalance Defects Diagnosis of Induction Machine Using Thermography and Machine Learning Algorithms

“…In cases where there is a tie in probability between two groups, the one with the highest probability is selected. It is postulated that the feature probabilities adhere to a Gaussian distribution, as depicted in equation ( 10) [32], [41], [43], [45]. In this article, the performance of the GNB model is evaluated using the temperature distribution dataset on the IM stator surface and within the ROI.…”

“…Each piece of information is subsequently placed in the category to which it is most closely associated. However, GNB considers not only the distance from the mean, but also how it compares to the class variance when calculating this proximity [45].…”

“…Once a model has been trained, it can be validated against the validation dataset. Finally, the validated model is tested on the test dataset [45].…”