NETWORK INTRUSION DETECTION METHODS BASED ON MACHINE LEARNING

With the rapid development of Internet technology, traditional intrusion detection methods have limitations. Although machine learning technology provides new ideas for intrusion detection, its efficiency and accuracy in large-scale data environment still need to be optimized. This study aims to propose an intrusion detection method that combines feature selection with optimized machine learning algorithms to solve the problems of data redundancy and category imbalance, and to reduce the false alarm rate. Based on the UNSW-NB15 dataset, ANOVA, chi-square test and Gini coefficient are used for feature selection, combined with principal component analysis (PCA) dimensionality reduction technique. The model is constructed by algorithms such as logistic regression and random forest, and hyperparameter optimization is carried out using GridSearchCV, and data imbalance and outliers are handled by stratified sampling and RobustScaler. The experiments show that the balanced accuracy of the logistic regression model is 70%, and the accuracy of the random forest model is 67.33%. Feature selection significantly improves the model performance. The method proposed in this study demonstrates high efficiency and reliability in large-scale network data and provides a technical basis for the design of real-time intrusion detection systems.