Dice Similarity Based Gaussian Deep Recurrent Neural Learning for Classification and Prediction with Big Data Analytics

Abstract

Big data analytics is a process of gathering large volume of data and organizing the present and past events to predict future events. Analyzing such a huge volume of data is not a simple task. Therefore, processing large data is a challenging one to predict an accurate event. The conventional techniques handling the large volume of data but the accurate prediction was not obtained since it failed to progressively learn the higher level features from raw inputs. An efficient Dice similarity based Gaussian Deep Recurrent Neural Learning Classifier (DS-GDRNLC) model is developed to enhance the prediction performance in terms of prediction time, prediction accuracy with big data. Initially, DS-GDRNLC model gathers huge volume of data from the big dataset (DS). After that, the gathered data are trained with several layers such as input layer, two hidden layers and output layer. The numbers of data are given to the input layer for performing the classification. Then the proposed DS-GDRNLC model uses two hidden layers to repeatedly learn the input data using a regression function. The regression function uses the dice similarity coefficient to find the relationship between the data and the predicted class. Then the analyzed results at the hidden layers are fed into the output layer. The Gaussian activation function is used at the output layer to verify the similarity value and mean of class. If the similarity value is closer to the mean of class, then the data are classified into that specific class. In this way, all the input data are accurately classified into the different classes resulting improves the Prediction Accuracy (PA). Finally, the training error rate is calculated for each classification results for obtaining the higher PA. This process repeated until the minimum error is obtained. Experimental evaluation is performed with big DS using different metrics such as PA, precision, recall, F-measure and Prediction Time (PT). The observed results confirm that the DS-GDRNLC model efficiently increases the PA, precision, recall as well as F-measure and minimizes the PT than the state-of-the-art methods.