Research On IoT Intrusion Detection Technology Based On Machine Learning

The rapid development of the Internet of Things(Io T)in recent years has led to its application in all areas of society.However,the long-term lack of human intervention in Io T devices and the varying levels of device technology have made Io T a major disaster area for vulnerability attacks.Intrusion detection technology,as an important branch of Io T security defence,can proactively detect intrusions in the Io T and provide timely warnings.With the development of artificial intelligence and the application of machine learning in the field of Io T intrusion detection,the capability of Io T intrusion detection has been further improved.However,the data features used to train Io T intrusion detection models are heterogeneous and redundant,which can affect the generalisation capability of existing intrusion detection models.In addition,the unbalanced distribution of data in the training set leads to extremely low detection rates of existing models for the classes with a few samples.To address the above problems,this paper conducts research in two aspects,namely feature extraction and data imbalance processing,and proposes a new Io T intrusion detection model based on machine learning,which effectively improves the overall detection rate and the detection rate of a few classes.The main research work of this paper is as follows.(1)To address the feature extraction problem,this paper proposes a feature extraction algorithm based on S-DBN,which uses a Dropout strategy to suppress the synergy of neurons in the same layer during the unsupervised process to improve the feature extraction ability,and introduces a penalty term in the training objective function using regularized optimization methods to reduce smaller redundant weights to prevent overfitting during the training of the S-DBN model.Experiments show that the S-DBN feature extraction algorithm can abstract high-dimensional data into lower-dimensional features that are easier to classify than traditional DBN and PCA feature extraction algorithms.(2)To address the data set imbalance problem,this paper proposes a data imbalance classification algorithm based on I-DQN,which formulates the data imbalance classification model as a sequential decision process and implements it through a DQN network.In the IDQN algorithm,the interaction rules and payoff functions are designed to make the I-DQN algorithm more sensitive to minority classes,thus increasing its ability to classify minority classes.Experiments show that the I-DQN algorithm improves the accuracy of minority classes to a higher level compared to other data imbalance processing methods.(3)Combining the research on feature extraction and data imbalance processing,this paper proposes an Io T intrusion detection model based on S-DBN and I-DQN,which consists of three modules: data preprocessing,S-DBN feature extraction and I-DQN data imbalance classification.The data preprocessing module is the numerical and normalization of the features of the original dataset.The S-DBN feature extraction module is to perform lowdimensional abstraction of the preprocessed dataset.The I-DQN data imbalance classification module is to classify the extracted features.The model is validated using the Bo T-Io T dataset for Io T intrusion detection.The experiments show that the proposed Io T intrusion detection model has higher detection accuracy in multiple classifications compared to other machine learning models,especially for the classes with a few samples,such as OS Fingerprinting,Data exfiltration,and keylogging.