DATA SECURITY DETECTION BASED ON IMPROVED PCA AND BP NEURAL NETWORK
Sugashini. K1, Kavya. S.P2
PG Student, Department of Computer Science and Engineering
Sri Shakthi Institute of Engineering and Technology, Coimbatore, India1
Assistant Professor, Department of Computer Science and Engineering
Sri Shakthi Institute of Engineering and Technology, Coimbatore, India2
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - With the growth of the Internet, digital attacks are evolving at a rapid pace, and the network security situation is far from ideal. For network examination of interruption identification, this paper uses AI (ML) and Deep Learning (DL) methodologies and provides a brief instructive exercise for each ML/DL strategy. Papers addressing each technique were listed, read, and summarized based on their tenuous or warm ties. Because data is so important in ML/DL methods, they depict a few of the most commonly used organization datasets in ML/DL, assess the challenges of using ML/DL for network security, and offer suggestions for further research. In the investigation of Intrusion Detection methods, KDD informative gathering is an important benchmark. A great deal of work is still being done to improve interruption recognition techniques, and research into the information used to prepare and test the location model is also a top priority, because higher information quality can improve disconnected interruption detection. This project investigates KDD informational collecting in four categories: Basic, Content, Traffic, and Host, in which all information ascribes can be organized using Modified Random Forest (MRF). The investigation on two unmistakable assessment measurements for an Intrusion Detection System, Detection Rate (DR) and False Alarm Rate (FAR), is complete (IDS). The commitment of each of four classes of attributes on DR and FAR is demonstrated as a result of this detailed inquiry into the informational index, which can aid in improving the appropriateness of the informational index to achieve the most extreme DR with the least FAR. The preliminary findings revealed that the proposed technique was able to achieve 88 percent accuracy with combining PCA and improved BP Neural Network, while comparing with other LSTM models such as LSTM, LSTM-PCS, PCS-BP Neural Network, also it produces less training time/s 23.2.
Key Words:Intrusion Detection System, Back propagation, Cyber Security
