Embedded Files
Research Work
Title : A Mixed Deep Learning and Statistical Approach for Network Anomaly Detection
Abstract :
Anomaly detection refers to the problem of identification data patterns, events in the The normal flow of network which raises suspicions by differing from the majority of data, which can help to prevent major attacks on an organization, but it is a tedious task to prepare a model that performs quite good in a zero-day attack situation, getting the anomaly containing data required to train model is very a complex job because each attack and anomaly vary as per the situation. Deep learning model-based approach uses freely available data and capture the data by tapping the network and acquire a .pcap file which can be used to extract features using a python based script then compute probabilities of each packet concerning Gaussian distribution because every normal packet will lie in normal distribution and packet containing anomaly will deviate from the normal distribution. The features along with their mapping values extracted from raw packet capture file in CSV format, to act as an input tensor to the deep-learning model and hidden layers contain a linear function with ReLU as activation function and loss criterion as binary entropy loss with sigmoid, Output of the model provides probabilities of the normal and anomalous behavior of the packet. Accuracy score calculation of results uses a joint probability-based approach, which assumes all the features are independent and it verifies the anomaly by comparing probabilities to a certain threshold. The Multilayer Perceptron model is trained using probabilities from statistical joint probabilities as ground truth for accuracy.
Title : Self-Evolving Deep Artificial Immune System for IoT Network Ecosystems
Abstract :
The Artificial Immune Systems The involvement of IoT in daily life is growing rapidly in different aspects
of human interaction, as this includes a significantly large amount of data exchange between a large number of devices the risk of data security is very high. In order to improve the security mechanism of such a dynamic and energy-efficient ecosystem, this paper proposes an artificial immune system-based approach that is inspired by natural immune systems that work with high precision to secure the network from foreign intrusions. Instead of mimicking the behavior of different cells that operate on the interaction of foreign objects, the current work explores the level-wise immunity mechanisms, which provides an efficient approach for detection of intrusion as well as satisfying complexity requirement. The proposed artificial immune system is designed to provide innate immunity to the IoT ecosystem with the help of statistical analysis of raw normal data flow in the network and to generate an innate immune response for an intrusion as the first layer of defense and an acquired immunity mechanism to learn the behavior of attack, memorize and respond more efficiently up-on the subsequent interaction of intrusion with the network as a secondary layer of defense. This work simulates the IoT network environment using the MQTTset dataset, which is more relevant for IoT networks instead of using benchmarks dataset like, KDDCUP99 and implements the immune system mechanisms using data-based statistical analysis and deep-learning approaches. The self-adaptive immune system yields better results than computational algorithm-based models as well as modern machine learning-based intrusion detection systems. Results conclude 99.87% accuracy on detecting the attack label on the MQTTset dataset with a 96.28% true positive rate and 97.5% true positive rate on real-time traffic.
Title : Vehicular Ad Hoc Networks: A comparative analysis of Security attacks and Privacy Requirements
Abstract :
The Artificial Immune Systems The involvement of IoT in daily life is growing rapidly in different aspects
of human interaction, as this includes a significantly large amount of data exchange between a large number of devices the risk of data security is very high. In order to improve the security mechanism of such a dynamic and energy-efficient ecosystem, this paper proposes an artificial immune system-based approach that is inspired by natural immune systems that work with high precision to secure the network from foreign intrusions. Instead of mimicking the behavior of different cells that operate on the interaction of foreign objects, the current work explores the level-wise immunity mechanisms, which provides an efficient approach for detection of intrusion as well as satisfying complexity requirement. The proposed artificial immune system is designed to provide innate immunity to the IoT ecosystem with the help of statistical analysis of raw normal data flow in the network and to generate an innate immune response for an intrusion as the first layer of defense and an acquired immunity mechanism to learn the behavior of attack, memorize and respond more efficiently up-on the subsequent interaction of intrusion with the network as a secondary layer of defense. This work simulates the IoT network environment using the MQTTset dataset, which is more relevant for IoT networks instead of using benchmarks dataset like, KDDCUP99 and implements the immune system mechanisms using data-based statistical analysis and deep-learning approaches. The self-adaptive immune system yields better results than computational algorithm-based models as well as modern machine learning-based intrusion detection systems. Results conclude 99.87% accuracy on detecting the attack label on the MQTTset dataset with a 96.28% true positive rate and 97.5% true positive rate on real-time traffic.
Google Sites
Report abuse