Automated Network Attack Detection Techniques Based on Improved Random Forests

The current network environment needs to face massive data and information; to ensure the network security, the automated detection technology of network attacks needs to be strengthened urgently.Therefore, the study adopts an improved random forest model based on Spark big data platform to optimize the detection process through online analysis and offline feedback mechanism.The focus is on the machine learning detector constructed by utilizing the MLlib library, and the semi-voting mechanism is used to speed up the model prediction.The experiments involve weight calculation methods weighted by correlation coefficients and enhance the model generalization ability by means of combinatorial optimization problems.The study also calculates decision tree similarity for random forest algorithm optimization in conjunction with packet characterization.In the experiments, on the UNSW-NB15 dataset, the improved model achieved a detection accuracy of 0.68 when using correlation coefficient weighting and tended to be stable with 16 decision trees.On the CICIDS2017 dataset, the detection accuracy obtained by this weighting method was 0.73 and stabilized with 12 decision trees.The Relative-RF-50% model using the semi-voting mechanism improved the prediction accuracy to 0.93844 on the CICIDS2017 dataset and obtained a substantial improvement in the execution time.Results show that the improved random forest model enhances the performance of automated cyber-attack detection, especially in terms of accuracy, recall, and efficiency showing obvious advantages.