This study focuses on enhancing the performance of intrusion detection systems (IDS) by proposing a hybrid model that combines multilayer neural networks with a dense-sparse-dense (DSD) training methodology using the UNSW-NB15 dataset. Various neural network architectures, including RNN, LSTM, and GRU, were evaluated to improve metrics such as accuracy and detection rates while addressing challenges posed by malware's increasing sophistication. The paper outlines the structure and implementation of the proposed hybrid model and discusses the training procedures, ultimately aiming for more accurate detection of network intrusions.