Enhancing Machine Learning Prediction in Cybersecurity Using Dynamic Feature Selector

Machine learning algorithms are becoming very efficient in intrusion detection systems with their real time response and adaptive process. A robust machine model can be deployed for anomaly by using a comprehensive dataset multiple attack types. Nowadays datasets contain many attributes. Such high dimensionality of poses significant challenge to information extraction terms space complexity. Moreover, having so attributes may hindrance towards creation decision boundary due noise the dataset. Large scale data redundant or insignificant features increases computational often decreases goodness fit which is critical issue cybersecurity. In this research, we have proposed implemented an feature selection algorithm filter variables. Our Dynamic Feature Selector (DFS) uses statistical analysis importance tests reduce complexity improve prediction accuracy. To evaluate DFS, conducted experiments on two used cybersecurity research namely Network Security Laboratory (NSL-KDD) University New South Wales (UNSW-NB15). meta-learning stage, four were compared Bidirectional Long Short-Term Memory (Bi-LSTM), Gated Recurrent Units, Random Forest Convolutional Neural (CNN-LSTM) accuracy estimation. For NSL-KDD, revealed increment from 99.54% 99.64% while reducing size one-hot encoded 123 50. UNSW-NB15 observed increase 90.98% 92.46% 196 47. The approach thus able achieve higher significantly lowering number required processing.