Original title: An LSTM-based intrusion detection model utilizing feature selection and dropout
Qi Wu, Yongqing He, Yang Liu, Ziyin Wang +4 · September 2026
Cyberattacks threaten both people and organizations, but detecting them is difficult. Network traffic data often contains redundant, unhelpful details. This extra noise leads to low detection accuracy and frequent false alarms. To fix this, researchers built a new intrusion detection model. First, it uses an algorithm called XGBoostAn algorithm that sorts through many features and picks the ones most useful for making predictions. to pick out the most relevant data features and cut out the clutter. Next, the system cleans the chosen data and sends it to a Long Short-Term Memory network, an artificial intelligence model that learns patterns over time. The team also used a technique called dropout to stop the network from memorizing the data too closely. In tests, this system beat older methods at spotting attacks. It proved especially good at catching rare threats, specifically Bot and Web attack intrusions.
You are a security guard watching a crowded street. With so many innocent details, you constantly mistake normal behavior for threats and miss real dangers. You decide to focus only on the few cues that truly matter, like quick hand movements, and you also recall what happened moments before. In this paper, the system does the same: it first selects the key features from network traffic, then uses a model that remembers sequences, so it spots real attacks without being swamped by background noise.
Network intrusions threaten both individuals and organizations, and traditional models often miss rare attack types. This model improves detection accuracy while lowering false alarms, and it catches minority attacks like Web and Bot that others overlook. That means better protection against unusual but dangerous threats.
The abstract doesn't state limitations, but it only tests on CICIDS2017, so performance on other datasets is unknown. It also compares only against CNN, Logistic Regression, and Bayesian models, not other modern approaches.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
An LSTM-based intrusion detection model utilizing feature selection and dropout — Qi Wu, Yongqing He, Yang Liu et al. — PeerJ Computer Science — doi:10.7717/peerj-cs.4057
Abstract licence: CC BY
Explanation generated by Synthesis from the abstract — it is not the original paper.