Linear Stochastic Feature Embedding based Regressive MIL Boost Data Classification for Streaming Data

Authors

  • Thangam M., Bhuvaneswari A.

Abstract

With the extensive growth of the Internet and rapid development, a large volume of data appears in different formats that are not able to process by conventional technologies. Since the dimensionality of the data stream is higher and the distribution of the data changes over time leads to improve accuracy in classification. Data Stream classification is the process of identifying the useful patterns from the continuous data records. In order to improve the classification accuracy of data stream, a novel technique called Linear Stochastic Feature Embedding Based Regressive Multiple Instance Learning (MIL) Boost Classification (LSFE-RMILBC) is introduced. The LSFE-RMILBC technique comprises of two major processes like dimensionality reduction and data classification. The dimensionality of the dataset is minimized by selecting the relevant features and removes the other features using the Modified Locally-Linear Stochastic Embedding technique. With the selected feature subset, the data stream classification is performed to improve the data stream classification. During the classification, the proposed MIL Boost Classification technique resolves the problem of concept drift with increase in classification accuracy and minimizes the false positive rate. MIL Boost Classification uses the Moving average regressive quadratic discriminant analysis as a weak learner for identifying the concept drift and for classifying the data with higher accuracy.   The extensive experimentation is conducted using weather data streams shows that the novel LSFE-RMILBC technique outperforms the recent methods. Specifically, the method yields higher classification accuracy, lesser time as well as lower false positive rates.                                           

Published

2020-12-30

Issue

Section

Articles