A case study for performance analysis of big data stream classification using spark architecture

• Published in: International journal of system assurance engineering and management 2024, Vol.15 (1), p.253-266
• Main Authors: Srivani, B., Sandhya, N., Rani, B. Padmaja
• Format: Article
• Language: English
• Subjects: Algorithms; Big Data; Case studies; Classification; Data analysis; Data collection; Data mining; Data transmission; Engineering; Engineering Economics; Feature selection; Logistics; Machine learning; Marketing; Optimization; Organization; Original Article; Parameter sensitivity; Performance evaluation; Quality Control; Reliability; Safety and Risk
• ISSN: 0975-6809; 0976-4348
• DOI: 10.1007/s13198-022-01703-4

A variety of huge data is being produced at an incredibly high speed in different sectors. Due to the large location of computing devices, the large volume of information is increasingly growing in the recent decades. A main role of big data is that a large set of data enables the machine learning techniques to obtain more accurate and better results. As the amount of data is exploding, it raises more challenges and opportunities for data analytic research in the data mining domain. The massively parallel databases not only have storage mechanisms but also have compute platforms. The extra capacity in the databases to really put some algorithms and move the data into in-memory to solve the problems. However, the big data stream contains different characteristics, such as high dimensionality, sparsity, volume and velocity. These characteristic features pose huge issues for the classification process when employing traditional data stream classification methods. For huge collection of data, effectively selecting the features and then classifying the data is important to make patterns. Recent feature selection strategies are involving the use of optimization methods for picking a subset of important features to get good classification results. Therefore, in this case study the feature selection is performed based on the Dragonfly Moth Search (DMS) optimization. The performance of the classification method is carried out in two different phases, such as offline and online phase by considering the master and slave node with stacked auto encoder (SAE) in the spark architecture. The parameters like accuracy, sensitivity and specificity metrics are evaluated on the performance of the DMS-SAE method.