Network intrusion systems are inevitable in protecting enterprise assets and improving cybersecurity. The research community is always on the lookout for new approaches to improve network intrusion detection. Network intrusion systems with deep learning models are the major advancement in this field. There are many varieties of network intrusion detection systems with deep learning models that are used nowadays. Even though researchers are investing heavily in their efforts on developing better network intrusion detection systems, rapid advancement in network intrusion attempts warrant further studies in this area. In this study, I am trying to explore the impact of the different most common feature selection methods on the performance of the LSTM-based network intrusion detection system. Benchmark network intrusion dataset UNSW-NB15 was explored for this study. This study explored 8 types of LSTM models in combination with different feature selection methods. The outcome of the study was very interesting as the LSTM model without applying any feature selection method, outperformed other combinations of models.