Abstract:
Compared to other conventional network threat prevention technologies, such as antivirus
and firewalls, network intrusion detection models (NIDS) offer a better option for network
intrusion detection. For intrusion detection, numerous machine learning algorithms have been
developed in the past. Dimensionality reduction-based techniques utilize the reconstruction error.
PCA is a well-known technique for dimensionality reduction. Consequently, it has been utilized
in recent articles for network intrusion detection. However, principal component analysis (PCA)
is a method for avoiding information loss while boosting interpretability by lowering the
dimensionality of such datasets. It accomplishes this by producing new, uncorrelated variables
that maximize variance one after the other. may result in several false reports as it can only
identify a linear association between features. But the majority of the features in network
intrusion detection are not linearly correlated. It has been confirmed that autoencoders are a more
effective dimensionality reduction technique than PCA for non-linearly correlated features.
In this research work, we propose an autoencoder-based approach for dimensionality
reduction. It constructively improves the prediction accuracy of a random forest classifier (RF).
The stacked autoencoder mechanism (SAE), an effective learning approach for building a new
feature representation unsupervised, is used to build the proposed model. The new attributes are
added to the RF algorithm after the pre-training step to boost the algorithm's classification
precision and intrusion detection ability. Additionally, the approach's effectiveness in binary and
multiclass classification is evaluated in comparison to that of a single RF classifier.
The NSL-KDD dataset, which includes normal and attack data, is used in the experimental
work. The result shows that the SAE-RF approach scores better result than that of a single RF.
SAE-RF scores 99% accuracy,99.48% F1-score,99.5 precision, and 99.48% recall for binary
class classification based on training data.
KeyWords:- Network intrusion detection , NSL-KDD , SAE-RF, RF ,PCA