IOT AWARE AUTOENCODER DEEP CAPSULE NETWORK FOR EFFICIENT PREDICTIVE ANALYTICS IN CLOUD

Abstract

Cloud computing is a term used to portray the delivery of on-demand computing resources such as servers, storage, databases, software, and analytics through the internet. Cloud computing facilitates organizations to access and store information without handling their own physical devices. In cloud computing model, predictive analytics is the process of forecasting the future outcomes or trends using historical data and machine learning algorithms. Many existing models have achieved lesser accuracy in predicative analytics model and increased their complexity due to variations in key features across different regions.  A novel Polynomial Regressive Robust Autoencoded  Deep Capsule Network (PRRA-CapsNet) model is developed for efficient predictive analytics in cloud computing. The main aim of PRRA-CapsNet model is to improve the accuracy of prediction with minimum time. The proposed deep capsule network includes the five different layers, namely one input layer, one output layer and three hidden layers namely convolutional layer, primary capsule network and class capsule network. Initially, IoT sensors are employed to gather the data points from different location. After that, the collected data points are transmitted to the input layer. The input layer transmitted the data points to convolutional layer. In the convolutional layer, PRRA-CapsNet model performs data preprocessing which includes missing data handling and outlier removal. Then, the pre-processed data points are transmitted to the primary capsule network. In that layer, Robust Autoencoder Feature Selection is employed in PRRA-CapsNet model for addressing the dimensionality reduction issues to identify the most important features in a dataset.  The relevant features are transmitted to the class capsule network. In that layer, the Kulczynski similarity function is used in PRRA-CapsNet model for performing the data classification in cloud environment. This in turn, efficient data analytics is carried out in cloud environment. Experimental analysis is carried out with the performance metrics like classification accuracy, classification time, precision, recall, f-measure and false positive rate with respect to number of data points.