IMPROVED ERROR REDUCED EXTREME LEARNING MACHINE (IERELM) CLASSIFIER FOR BIG DATA ANALYTICS

摘要

Nowadays the term "Big Data Analytics" has been talked everywhere due to the advancement of information technology, evolution of computer applications, mobile communications and much more. This paves the way for doing lot of research in this big data arena. It is common to known that classification is one among the thrust research dimension in the field of big data particularly in the field of data analytics. Machine learning algorithms have lot of scope on analytics; particularly extreme learning machine is a kind of feed-forward neural network that is commonly applied for performing classification task. This machine learning ELM algorithm has single layer hidden nodes with randomly assigned weights for connecting with hidden layer. It is to be noted that feed-forward neural networks in extreme learning machine are poor in updating the weights that leads to performance degradation. Computational complexity is certainly more when applying ELM for big data analytics. This part of doctoral research work inclines high motivation for improving the performance of ELM in terms of reducing the error and we named it as Improved Error Reduced ELM shortly coined as IERELM. The performance of the proposed IERELM mechanism is applied for performing the classification task in KDD Cup 99 multivariate dataset that contains 40,00,020 instances with 42 attributes. Obtained results portrays that the proposed IERELM performs better in terms of detection rate, false alarm rate and elapsed time to perform classification.