An Intrusion Detection Method Based On Feature Selection And Machine Learning Algorithm

Io T(Internet of Things)devices are being used in a wide range of applications,including Smart Home,Smart Schools,Smart Healthcare,and many more.The system's intelligence,power efficiency,and data processing capabilities are enhanced by the use of Io T devices,but the system's security is put at danger.A system built on the Internet of Things(Io T)is vulnerable if it lacks adequate security.It is the application layer of Io T devices that an attacker focuses on.Distributed denial of service attack(DDos),Satan Attack,Spyware attacks,and other sorts of attacks are posing a security risk to Io T-based devices.To detect assaults in an Io T-based system,a number of previous studies have been conducted using traffic data from Io T devices.As part of identifying assaults,source and destination bytes play a significant role.In this thesis,the main research goals which are the detection and prediction about DDos attack focus on Io T devices.At first,the data preprocessing methods are used,such as data balance,outliers elimination etc.The second,a vulnerabilities detection system was built,which based on the original and target bytes as well as other parameters about the flow characteristics of Io T equipment's,the last,an Io T devices comprehensive attack classification method based on machine learning is proposed.According to this study,The contribution of this thesis are the following:(1)The major objective is to demonstrate the security and efficacy of a machine learning-based system for predicting security threats in Io T devices on the application layer using traffic data.(2)Furthermore,the models should be evaluated on the basis of their performance criteria.With skewed data,prediction systems encounter difficulties.So in order to ensure that the dataset was balanced,data balancing procedures had to be employed(Tomek and SMOTE).(3)To facilitate analysis,outliers were removed from the data.This study also looked at categorization models based on machine learning in the hopes of spotting an attack as early as possible.The dataset was balanced in order to assess classifier accuracy between models.Soft and hard voting classifiers were 90%and 89.999% accurate in detecting DDo S assaults,respectively.