Analysis of Data Mining Techniques on KDD-Cup'99, NSL-KDD and UNSW-NB15 Datasets for Intrusion Detection.

Intrusion detection is one of the important fields that can detect abnormal behavior on the network. Intrusion detection systems are expected to grow in the market, and the demand for these systems will increase soon. To find the anomalous behavior on the network, building models using data mining classifiers such as Random Forest (RF), K-Nearest Neighborhood (KNN), and Naïve Bayes (NB) is used. The performance of the classifiers was tested on three different data sets (KDDCUP-99, UNSW-NB15, and NSL-KDD). The results proved that Random Forest is the most reliable and adaptable algorithm across all three datasets. Its ensemble learning capabilities, suitability for high-dimensional data, and resistance to overfitting make it a valuable choice for intrusion detection This research is limited in some areas. First, the number of data sets used in this research is limited to three; future studies can include more data sets to get better comparative results. The second limitation is that the technique used in this paper is the classification algorithms only. The researchers suggest using more data mining techniques and comparing the results. Neural Networks is one of the algorithms that can be used in this field. The researchers suggested that future studies focus more on the in-depth analysis of the intricacies of specific datasets, focusing on factors that impact algorithm performance, including data quality and complexity, which can provide insights into fine-tuning algorithms for better results.