Enhanced Gaining-Sharing Knowledge Optimization Algorithm for 3D Compression of Intrusion Detection Dataset

The study addresses challenges associated with intrusion detection datasets in terms of high dimensionality by adopting new methods to reduce their size and improve efficiency. Firstly, the dataset's features were reduced using a selection method based on Pearson Correlation, Entropy, and Information Gain (PC-E-IG) to identify only those with a high correlation to the target class. Secondly, filtering methods were applied to reduce the dataset's records based on two situations. The Gaining–Sharing Knowledge (GSK) optimization algorithm was employed, where its fitness function was replaced via re-combination methods, and the best method was chosen using the Ackley evaluation function. The first method reduced the size of the data by 20% by filtering a group of training data with different fitness values. The second method filtered and isolated records with the Normal target from the training and testing datasets, resulting in a 19.69% and 19.48% reduction in the size of the dataset, respectively. The study used the KDDCup99 dataset and Multilayer Perceptron to test the efficiency of the system and compare its results with the original algorithm, resulting in an impressive increase in model accuracy and a clear reduction in execution time. The study's first reduction method achieved an accuracy of 92.45% and an execution time of 0.09 s. The second reduction method resulted in an accuracy of 93.8% and an execution time of 0.07 s. The main benefit of these results is that they demonstrate the effectiveness of the study's methods in improving the accuracy of intrusion detection models while also reducing execution time.