Machine Learning-Based Adaptive Synthetic Sampling Technique for Intrusion Detection

Traditional firewalls and data encryption techniques can no longer match the demands of current IoT network security due to rising amount variety threats. In order manage risks, intrusion detection solutions have been advised. Even though machine learning (ML) helps widely used currently in use, these algorithms struggle with low rates requirement for extensive feature engineering. The deep model is a method traffic anomaly that suggested by this study. To extract sequence properties flow through CNN, it combines an attention mechanism Long Short Term Memory (LSTM) network. This uses adaptive synthetic sampling (ADASYN) increase size minority-class samples. proposed models demonstrated acceptable precision recall each class when binary-class classification, proving their stability capacity identify all classes correctly. MLP classifier’s accuracy, precision, recall, F1 value were 87%, 89%, respectively, AUC score 0.88. Overall, performed well. attack all-class exhibited good AUCs macro metrics, same as classifier, which had 83% 0.94. Additionally, trained classifier integrated ADAM optimizer category cross-entropy loss function classification. With 94%, possessed 84% 87% score. A further indication hybrid model’s ability combine benefits both improve overall performance was regularly outperformed model. accuracy are better than those earlier comparable algorithms, according experimental results using publicly accessible benchmark dataset (NSL–KDD).