This work presents the design of a limited-computational-resources prototype for cardiac arrhythmias detection. To do so, a heartbeat classification strategy is developed aimed at identifying normal and pathological heartbeats in long-term electrocardiographic (Holter) recordings. By incorporating an embedded system, a low computational cost system is developed, which is capable of analyzing the characteristics of QRS complexes waves -being representative waves of the heartbeat, and their analysis allows for the identification of ventricular arrhythmias. To develop this initial prototype, we experimentally demonstrated that the use of the algorithm of k nearest neighbors (k-NN) together with a stage for selection of training data variables is a good alternative, and represents a major contribution of this work. Experiments are performed on the Massachusetts Institute of Technology (MIT) cardiac arrhythmia database. Obtained results are satisfactory and promising.
Broadly, the area of dimensionality reduction (DR) is aimed at providing ways to harness high dimensional (HD) information through the generation of lower dimensional (LD) representations, by following a certain data-structure-preservation criterion. In literature there have been reported dozens of DR techniques, which are commonly used as a pre-processing stage withing exploratory data analyses for either machine learning or information visualization (IV) purposes. Nonetheless, the selection of a proper method is a nontrivial and -very often- toilsome task. In this sense, a readily and natural way to incorporate an expert's criterion into the analysis process, while making this task more tractable is the use of interactive IV approaches. Regarding the incorporation of experts' prior knowledge there still exists a range of open issues. In this work, we introduce a here-named Inverse Data Visualization Framework (IDVF), which is an initial approach to make the input prior knowledge directly interpretable. Our framework is based on 2D-scatter-plots visuals and spectral kernel-driven DR techniques. To capture either the user's knowledge or requirements, users are requested to provide changes or movements of data points in such a manner that resulting points are located where best convenient according to the user's criterion. Next, following a Kernel Principal Component Analysis approach and a mixture of kernel matrices, our framework accordingly estimates an approximate LD space. Then, the rationale behind the proposed IDVF is to adjust as accurate as possible the resulting LD space to the representation fulfilling users' knowledge and requirements. Results are greatly promising and open the possibility to novel DR-based visualizations approaches.
Time-varying data characterization and classification is a field of great interest in both scientific and technology communities. There exists a wide range of applications and challenging open issues such as: automatic motion segmentation, moving-object tracking, and movement forecasting, among others. In this paper, we study the use of the so-called kernel spectral clustering (KSC) approach to capture the dynamic behavior of frames -representing rotating objects- by means of kernel functions and feature relevance values. On the basis of previous research works, we formally derive a here-called tracking vector able to unveil sequential behavior patterns. As a remarkable outcome, we alternatively introduce an encoded version of the tracking vector by converting into decimal numbers the resulting clustering indicators. To evaluate our approach, we test the studied KSC-based tracking over a rotating object from the COIL 20 database. Preliminary results produce clear evidence about the relationship between the clustering indicators and the starting/ending time instance of a specific dynamic sequence.
This work presents a dimensionality reduction (DR) framework that enables users to perform either the selection or mixture of DR methods by means of an interactive model, here named Geo-Desic approach. Such a model consists of linear combination of kernel-based representations of DR methods, wherein the corresponding coefficients are related to coordinated latitude and longitude inside of the world map. By incorporating the Geo-Desic approach within an interface, the combination may be made easily and intuitively by users -even non-expert ones- fulfilling their criteria and needs, by just picking up points from the map. Experimental results demonstrates the usability and ability of DR methods representation of proposed approach.
Dimensionality reduction (DR) methods are able to produce low-dimensional representations of an input data sets which may become intelligible for human perception. Nonetheless, most existing DR approaches lack the ability to naturally provide the users with the faculty of controlability and interactivity. In this connection, data visualization (DataVis) results in an ideal complement. This work presents an integration of DR and DataVis through a new approach for data visualization based on a mixture of DR resultant representations while using visualization principle. Particularly, the mixture is done through a weighted sum, whose weighting factors are defined by the user through a novel interface. The interface’s concept relies on the combination of the color-based and geometrical perception in a circular framework so that the users may have a at hand several indicators (shape, color, surface size) to make a decision on a specific data representation. Besides, pairwise similarities are plotted as a non-weighted graph to include a graphic notion of the structure of input data. Therefore, the proposed visualization approach enables the user to interactively combine DR methods, while providing information about the structure of original data, making then the selection of a DR scheme more intuitive.
This work describes a new model for interactive data visualization followed from a dimensionality-reduction (DR)-based approach. Particularly, the mixture of the resultant spaces of DR methods is considered, which is carried out by a weighted sum. For the sake of user interaction, corresponding weighting factors are given through an intuitive color-based interface. Also, to depict the DR outcomes while showing information about the input high-dimensional data space, the low-dimensional representations reached by the mixture is graphically presented using scatter plots improved with an interactive data-driven visualization. In this connection, a constrained dissimilarity approach is proposed to define the graph to be drawn on the scatter plot. Proposed data visualization model enables users (even non-expert ones) to make decisions on what are the most suitable lower-dimensional representations of the original data in a friendly-user fashion.
This work presents an improved interactive data visualization interface based on a mixture of the outcomes of dimensionality reduction (DR) methods. Broadly, it works as follows: The user can input the mixture weighting factors through a visual and intuitive interface with a primary-light-colors-based model (Red, Green, and Blue). By design, such a mixture is a weighted sum of the color tone. Additionally, the low-dimensional representation space produced by DR methods are graphically depicted using scatter plots powered via an interactive data-driven visualization. To do so, pairwise similarities are calculated and employed to define the graph to simultaneously be drawn over the scatter plot. Our interface enables the user to interactively combine DR methods by the human perception of color, while providing information about the structure of original data. Then, it makes the selection of a DR scheme more intuitive -even for non-expert users.
This work presents a new interactive data visualization approach based on mixture of the outcomes of dimensionality reduction (DR) methods. Such a mixture is a weighted sum, whose weighting factors are defined by the user through a visual and intuitive interface. Additionally, the low-dimensional representation space produced by DR methods are graphically depicted using scatter plots powered via an interactive data-driven visualization. To do so, pairwise similarities are calculated and employed to define the graph to be drawn on the scatter plot. Our visualization approach enables the user to interactively combine DR methods while provided information about the structure of original data, making then the selection of a DR scheme more intuitive.