R. Pradeep Kumar

Histogram PCA

Histograms are data objects that are commonly used to characterize media objects like image, video, audio etc. Symbolic Data Analysis (SDA) is a field which deals with extracting knowledge and relationship from such complex data objects. The current research scenario of SDA has contributions related to dimensionality reduction of interval kind data. This paper makes an important attempt to analyze a symbolic data set for dimensionality reduction when the features are of histogram type. The result of an in-depth analysis of such a histogram data set has lead to proposing basic arithmetic and definitions related to histogram data. The basic arithmetic has been used for dimensionality reduction modeling of histogram data set through Histogram PCA. The modeling procedure is demonstrated by experiments with 700x3 data, iris data and 80X data. The utility/applicability of Histogram PCA is validated by clustering the above data.

An Approach Based on Regression Line Features for Low Complexity Content Based Image Retrieval

Similarity matching is one of the important tasks in content based image retrieval systems. Similarity matching involves the computation of distance between the feature vectors characterizing the image samples. Conventional techniques like pixel based similarity matching are computationally costly and time consuming. In recent years the tremendous increase in multi media databases, especially image databases calls for fast and efficient image retrieval mechanisms. Multiresolution based approaches through multiresolution histograms and wavelet histograms proposed recently are proven to be computationally efficient. In this paper, we propose a methodology based on regression line features for further reducing the computational complexity of these multiresolution histogram based techniques

Sequence Compulsive Incremental Updating of Knowledge in Learning Management Systems

Growing popularity of Learning Management Systems (LMS) coupled with setting up of variety of rubrics to evaluate methods of Learning, Teaching and Assessment Strategies (LTAS) by various accreditation boards has compelled many establishments/universities to run all their courses through one or the other forms of LMS. This has paved way to gather large amount of data on a day to day basis in an incremental way, making LMS data suitable for incremental learning through data mining techniques. The data mining technique which is employed in this research is clustering. This paper focuses on challenges involved in the instantaneous knowledge extraction from such an environment where streams of heterogeneous log records are generated every moment. In obtaining the overall knowledge from such LMS data, we have proposed a novel idea in which instead of reprocessing the entire data from the beginning, we processed only the recent chunk of data (incremental part) and append the obtained knowledge to the knowledge extracted from previous chunk(s). Obtained results when compared with teachers handling the modules/subjects match exactly with the expected results.

WaveSim transform for multi-channel signal data mining through linear regression PCA

Temporal data mining is concerned with the analysis of temporal data and finding temporal patterns, regularities, trends, clusters in sets of temporal data. In this paper we extract regression features from the coefficients obtained by applying WaveSim Transform on Multi-Channel signals. WaveSim Transform is a reverse approach for generating Wavelet Transform like coefficients by using a conventional similarity measure between the function f(t) and the wavelet. WaveSim transform provides a means to analyze a temporal data at multiple resolutions. We propose a method for computing principal components when the feature is of linear regression type i.e. a line. The resultant principal component features are also lines. So through PCA we achieve dimensionality reduction and thus we show that from the first few principal component regression lines we can achieve a good classification of the objects or samples. The techniques have been tested on an EEG dataset recorded through 64 channels and the results are very encouraging.