Enhancement of Accuracy and Performance of LSTM Based IDS System

In cybersecurity, IDS are quite crucial as they help to identify and lower probable dangers. However, particularly in the identification of complex attack patterns, standard IDS models might suffer from performance and accuracy limits. This study aims to boost computational economy and detection accuracy by presenting enhanced LSTM-based IDS employing hybrid DL models, feature selection techniques, and optimal hyperparameters. The study applies enhanced preprocessing, data augmentation, and ensemble learning to improve intrusion categorisation. Testing the suggested model on benchmark datasets over conventional IDS models yields significant gains in accuracy parameters. The study also looks at how network traffic characteristics impact detection performance and approaches to lower false positives. Based on low latency and high scalability and improved real-time detection capabilities, the proposed LSTM-based IDS seems to be a workable solution for present cybersecurity problems. Present research make use of optimized LSTM model that considers multiple hidden layer and it support optimization of epochs and batches along with learning rate. In proposed work during preprocessing phase, the missing value data has been eliminated. Then LSTM model has been used with optimized hyperparameters. The dense layer has been added to reduce the probability of overfitting. It has been observed that Batch size and number of epoch influences the performance as well as accuracy. Moreover, Eliminating incomplete values helps in minimizes overfitting/under fitting, directly improving prediction precision along with reliability and performance.