Özet:

Abstract Data mining as well as big data analytics represent approaches for analysing and extracting useful secret data. Although big data is complicated and large in volume, conventional methods to interpretation and retrieval do not function well. Data clustering is a common data mining approach that divides nodes into categories and makes it possible to retrieve features out of these groups. Conventional clustering techniques, including such k-means clustering as well as hierarchical clustering, are inefficient because the reliability of the groups they generate is harmed. As a result, an efficient and relatively extensible clustering technique is required. In this paper we proposed novel similarity-based clustering techniques on large unstructured transaction dataset. The HDFS file system log data has collected from real time Virtual Machine’s (VM’s) and generates the clusters, using reinforcement learning technique. Initially data has collected from various VM’s and proposed dimensionality reduction technique has used for data reduction. The Q-learning based reinforcement learning algorithm has applied on generated event. The activation function calculates the current weight for each transaction according to reward and penalty. Finally, the threshold-based entropy function generates a final cluster. After the clustering process modified Support vector Machine (mSVM) as supervised classifier has applied on entire label data.  In the extensive experimental analysis, we evaluate proposed model performance with existing techniques. The proposed clustering and classification method beats the comparable models in terms of average operating time and average clustered error, according to tests conducted on actual, synthetic, and automatically created datasets.

Anahtar Kelimeler: