Developing optimal solutions for the data management challenges in social internet of things

Abstract

The convergence of the Social Internet of Things (SIoT) and Big newlineData has presented a significant challenge in effectively classifying and newlineanalyzing massive amounts of generated electronic data. While several newlinemachine learning techniques are employed to extract useful information from newlinebig data, they face challenges such as high training time, memory volume, newlineand computation costs in performing the classification task. This thesis newlineproposes a comprehensive approach that incorporates two innovative newlinemethods: the Fuzzy Optimized Deep Convolutional Neural Network newline(FDCNN) and the Marine Predator-based Deep Recurrent Neural Network newline(MDRNN) to address these challenges and enhance classification accuracy newlinewhile reducing energy consumption. newlineThe FDCNN method integrates a deep convolutional neural newlinenetwork with the fuzzy-based Remora Optimization (BRO) algorithm for newlineSIoT big data classification. This combination allows for effective feature newlineselection and improved performance. On the other hand, the MDRNN newlinemethod leverages deep recurrent neural networks and the marine predator newlinealgorithm to classify SIoT big data. An adaptive filter is employed to select a newlinesuitable subset of data, eliminating unwanted noise and redundant newlineinformation. The Hadoop MapReduce framework is used to reduce the newlinedimension of the data, improving the performance of the proposed method. newlineThe modified relief technique is also utilized for optimal feature/attribute newlineselection, thereby enhancing classification accuracy. newline