Evaluation of IDS model by improving accuracy and reducing overfitting using stacking LSTM.

Utilizing Machine Learning (ML) techniques to construct models for Intrusion Detection Systems (IDS) has emerged as a leading strategy to enhance security measures. Previously, we applied an ensemble of Support Vector Machines (SVM) for model training, achieving an accuracy detection rate of 88.6% on familiar datasets. However, it performs poorly on unknown datasets due to overfitting problems. This work aims to improve accuracy and tackle overfitting problems by implementing stacked Long Short-Term Memory (LSTM) networks. Furthermore, analysis of variance (ANOVA) with F-test is used to find the important feature. The synthetic minority oversampling technique (SMOTE) algorithm is applied to obtain balanced data. Several experiments were conducted to analyze the performance of the proposed approach, including investigating stacked LSTM with tuning hyperparameters. The proposed method achieved the highest prediction accuracy of 97.23% and an overfitting value of 0.24%.