Gaddam Keerthi, Tirunagiri Kaushik Prasad, Sudersan Behera
Keywords:
visual analytics, machine learning, data quality, multi-dimensional data visualization, user interaction.
Abstract
To train algorithm-specific models, machine learning algorithms and conventional data mining techniques typically need a lot of data, with little to no user input during the model-building phase. Sometimes, such a "big data'' based machine learning technique is impractical for use in settings like clinical trials where gathering or processing data is exceedingly costly or challenging. Furthermore, in some subjects, like the biological sciences, expert knowledge can be quite helpful when developing models. We present a novel technique for interactive machine learning and visual data mining using visual analytics in this research. This method uses multidimensional data visualization approaches to make it easier for users to participate in the mining and machine learning processes. This enhances the effectiveness of model creation by enabling dynamic user feedback in many ways, including data selection, data labeling, and data correction. This method can have a major influence on applications where obtaining huge amounts of data is difficult, as it can drastically reduce the amount of data needed to train an appropriate model. Two application problems—the handwriting recognition problem and the human cognitive score prediction problem—are used to evaluate the suggested methodology. The results of both experiments demonstrate that interactive machine learning and data mining aided by visualization can get the same accuracy as an automated process using far smaller training data sets.
Article Details
Unique Paper ID: 162646
Publication Volume & Issue: Volume 10, Issue 10
Page(s): 602 - 606
Article Preview & Download
Share This Article
Join our RMS
Conference Alert
NCSEM 2024
National Conference on Sustainable Engineering and Management - 2024