An automated internet of behavior detection method based on feature selection and multiple pooling using network data

Nowadays, the internet is the most used communication environment, and therefore it becomes very important to try to determine the behavior of users regarding internet use. Due to the internet of behaviors (IoBe) information, user-specific recommendations can be customized in various fields such as trade, health, economy, law, and entertainment. This study presents an automated and accurate classification model and a new dataset to detect IoBe. This model uses internet packets, and a dataset is created using variable behaviors. A new feature engineering model is presented to classify IoBe by using the collected packets. The developed model has three phases: feature increasing using four pooling functions/methods, ReliefF based meaningful feature selection, classification, and majority voting. The developed model has been tested on the collected IoBe dataset and CICDarknet2020 dataset to predict behaviors. The presented pooling increasing method and ReliefF-based model attained 83.01% and 93.90% accuracy for IoBe and CICDarknet2020 datasets. These classification accuracies and findings demonstrated the success of the proposed feature engineering model, and a new dataset has been publicly published to contribute IoBe works.