Weighted bidirectional gated recurrent network for event detection

Modern information technology is able to store enormous amounts of information even at high speeds and volumes. Meanwhile, handling continuous data streams becomes a complicated task and thus, a hybrid weighted recurrent neural network as well as bidirectional gated recurrent unit (hybrid WRBG) method is proposed for ideal feature sub-selection from the hyperspace of big data. Here, three datasets are utilized, namely as MAVEN dataset, Climate Change Twitter dataset, and event detection dataset to examine the proposed hybrid WRBG method. By utilizing the fuzzy elephant herding optimization (FEHO) which is a form of swarm search which delivers higher analytical accuracy within a practical processing time, the feature selection is specifically created for detection of events. Also, by attaining a tradeoff in the range of bias and variance terms, the classifier error is reduced through the FEHO algorithm. A weighted recurrent neural network (weighted RNN) ensemble with a bidirectional gated recurrent unit classifier is employed in order to automatically update current concepts in big data streams. The proposed model achieves an accuracy of 98.97%, a precision of 98.87%, an f-score of 98.72%, and a kappa score value is 0.92.