Abstract: In more recent times, there has been an increase in the number of people using computers, as a result of which there is widespread use of the Internet. The use of the Internet enables hackers to access computers using new, more sophisticated, and more complex forms of attacks, to protect computers from them Intrusion Detection System (IDS) is employed, which has been trained using a number of machine learning techniques as well as datasets. In some networks, the datasets used are acquired over time and usually contain up-to-date data. Furthermore, they are imbalanced and unable to store enough data to withstand all types of attacks. The efficiency of current IDSs is harmed by these inconsistencies and out dated datasets, especially for attacks that are infrequently encountered. We propose a machine learning-based IDSs in this paper, using K-Nearest Neighbour, Decision-Tree, SVM, LSTM, and SMOTE algorithms. To make IDS more logical, an up-to-date security database, CSE-CIC-IDS2018, can be used in place of older and more widely used datasets. The selected database is also not balanced. As a result, utilizing a data model known as the Synthetic Minority Oversampling technique (SMOTE), the rate of inequality in the dataset is lowered to improve the reliability of the system and to avoid inconsistent access and false alarms, a mechanism based on the types of attacks was developed. Data is processed in small classes, and their numbers grow to medium data size in this fashion. The proposed strategy considerably boosts the detection rate of attacks that are infrequently encountered, according to experimental results.
Keywords: IDS, intrusion detection, SVM, LSTM, SMOTE, machine-learning, CSE-CIC-IDS2018.
| DOI: 10.17148/IJARCCE.2021.10591