Ahsan Abdullah

A new biclustering technique based on crossing minimization

Abstract Clustering only the records in a database (or data matrix) gives a global view of the data. For a detailed analysis or a local view, biclustering or co-clustering is required, involving the clustering of the records and the attributes simultaneously. In this paper, a new graph-drawing-based biclustering technique is proposed based on the crossing minimization paradigm that is shown to work for asymmetric overlapping biclusters in the presence of noise. Both simulated and real world data sets are used to demonstrate the superior performance of the new technique compared with two other conventional biclustering approaches.

A Cognitively Inspired Approach to Two-Way Cluster Extraction from One-Way Clustered Data

Cluster extraction is a vital part of data mining; however, humans and computers perform it very differently. Humans tend to estimate, perceive or visualize clusters cognitively, while digital computers either perform an exact extraction, follow a fuzzy approach, or organize the clusters in a hierarchical tree. In real data sets, the clusters are not only of different densities, but have embedded noise and are nested, thus making their extraction more challenging. In this paper, we propose a density-based technique for extracting connected rectangular clusters that may go undetected by traditional cluster extraction techniques. The proposed technique is inspired by the human cognition approach of appropriately scaling the level of detail, by going from low level of detail, i.e., one-way clustering to high level of detail, i.e., biclustering, in the dimension of interest, as in online analytical processing. A number of experiments were performed using simulated and real data sets and comparison of the proposed technique made with four popular cluster extraction techniques (DBSCAN, CLIQUE, k-medoids and k-means) with promising results.

Biclustering gene expression data in the presence of noise

Production of gene expression chip involves a large number of error-prone steps that lead to a high level of noise in the corresponding data. Given the variety of available biclustering algorithms, one of the problems faced by biologists is the selection of the algorithm most appropriate for a given gene expression data set. This paper compares two techniques for biclustering of gene expression data i.e. a recent technique based on crossing minimization paradigm and the other being Order Preserving Sub Matrix (OPSM) technique. The main parameter for evaluation being the quality of the results in the presence of noise in gene expression data. The evaluation is based on using simulated data as well as real data. Several limitations of OPSM were exposed during the analysis, the key being its susceptibility to noise.

Automatic Text Removal and Replacement in Scanned Meteorological Thematic Maps

Geo-referenced data is required in many Geographic Information System (GIS) applications and one source of such data being maps. Meteorological maps are color-coded with different regions corresponding to different values of a parameter, parsing the image to convert into data is not very difficult. However, text and different planimetric elements overlaid on the regions in the map makes accurate image to data conversion a challenging problem, because it is almost impossible to exactly replace what was underneath the text or icons i.e., the need for inpainting. In this study, we propose a probabilistic technique that uses the probability of occurrence of colors present in the map along with the occurrence of those colors in the spatial neighborhood of the pixel (corresponding to text) whose color is required to be replaced. We test the limits of our proposed technique using simulated data and compare results with a popular image editing tool using public domain data with promising results.

A Novel Near-Infrared Spectroscopy Based Spatiotemporal Cognition Study of the Human Brain Using Clustering

In this study, we investigate how the two hemispheres of the brain are involved spatiotemporally in a cognitive-based setup when people relate different colors with different concepts (for example, the color ‘blue’ associated with the word ‘dependable’ or ‘cheap’) objectively or subjectively. We developed an experimental setup using a 17-channel near-infrared spectroscopy (NIRS) device to measure the changes in brain hemoglobin concentration during a concept–color association task in a block design paradigm. The channel-wise activation data were recorded for 10 male students; after cleansing, the data were clustered using an indigenous clustering technique to identify channels having similar spatiotemporal activity. Data mining was imperative because of the big data generated by NIRS (ca. 0.1+ MB textual data captured per sec involving high volume and veracity), for which the traditional statistical techniques for data analysis could have failed to discover the patterns of interest. The results showed that it was possible to associate brain activities in the two hemispheres to study the association among linguistic concepts and colors, with most neural activity taking place in the right hemisphere of the brain characterized with intuition, subjectivity, etc. Thus, the study suggests novel application areas of neural activity analysis, such as color as marketing cue, response of obese versus lean to food intake, traditional versus neural data validation.

Introduction: Dealing with Big Data-Lessons from Cognitive Computing

Big Data analytics is an emerging area of fast-growing importance as it provides innovative ways to efficiently analyse large and complex data effectively. Many approaches have been developed to extract meaningful information and knowledge from large multidimensional data sets involving high veracity, volume, velocity, and value. Unlike traditional approaches, cognitive computing systems are not based on predetermined answers or actions to perform Big Data analytics; instead, they are trained using efficient, scalable, natural, and biologically inspired approaches, including computational intelligence and machine learning algorithms without compromising their sophistication and performance. Cognitive computing systems interact with, learn naturally from people and Big data, and thus help increase the productivity of what either humans or machines could do on their own. This Special Issue is aimed at promoting multidisciplinary research, particularly in the complementary areas of cognitive psychology, biology, and computing science for dealing with Big Data challenges across these domains. Accordingly, the articles selected in this Special Issue report a range of cognitively inspired computational approaches to deal with different aspects of Big Data analytics. Following a rigorous peer-review process, ranging from three-to-four rounds of revisions; five articles have been accepted for publication. The reviews of the final article co-authored by the guest editors were handled independently by another editorial board member. In the first article presented in this Special Issue, Wu, Pang, and Coghill propose an integrative qualitative and quantitative modelling framework for inferring biochemical systems that could help, especially biologists, better understand natural biochemical systems. The proposed framework demonstrates how the identification of biochemical systems can be performed and evolved in an integrative manner by reusing, composing, and evolving biochemical modules qualitatively, and by mutating kinetic rates quantitatively. In the next article, Du et al. propose a novel computational method to detect specific biomarkers for different groups of cancer types. The proposed method identifies specific biomarkers for a given cancer group based on different factors, including the same survival rates, the same type, and grade. The proposed methodology is thoroughly evaluated using eight cancer types and a number of benchmark microarray gene expression data sets from public databases. In the next article, Lin et al. describe a cognitively inspired psycho-linguistically motivated approach to build a quiz generation system, which combines similarity information derived from linked data and Resource Description Framework (RDF) resources. The authors effectively exploit this similarity information to gauge the difficulty of quizzes, confirming that a higher similarity among concepts leads to a more difficult quiz formulation, and vice versa. The proposed methodology is evaluated & Ahsan Abdullah aabdullah1@kau.edu.sa

A Methodology and a Tool to Prepare Agro-Meteorological Maps as a Source of Big Data

The volumetric aspect of big data is increasing because of introduction of new methods to generate-capture data and the diverse needs of that data. Agro-meteorological maps (e.g. Digital maps, printed maps, scanned maps etc.) and satellite imagery are also sources of big data in terms of volume, variety in format, scale, representation, then presence of noise resulting in veracity and the velocity in terms of rate of availability of satellite imagery. One main problem with these domain specific agro-meteorological maps is presence of veracity in terms of noise (i.e. Irrelevant data in maps) such as, city names, symbols, administrative boundaries etc. If we can successfully remove noise from these maps and replace with right data, we can convert these maps into rich sources of digital data resulting in new opportunities in traditional agricultural sector, in policy making and decision support. In this paper we present an approach and its implementation that can be used to minimize the veracity in maps by cleaning the noise from maps and replacing the noise with right data. Once maps are cleaned, right data can be extracted from them. This data can be used by different big data applications and knowledge based systems in decision and policy making in different sectors.

Digital Divide and Caste in Rural Pakistan

This study examined how caste influences adoption of information and communication technologies. It entailed a survey of 2750 farmers, who were interviewed in person, in rural Punjab (Pakistan), and two different methods for data analysis—data mining and indices-based statistical analysis. The analysis shows digital divides at the caste level. Furthermore, it shows that old and new technologies diffuse differently among castes.

Analysis of pesticide application practices using an intelligent Agriculture Decision Support System (ADSS)

Pesticides are used for controlling pests, but at the same time they have impacts on the environment as well as the product itself. Although cotton covers 2.5% of the world’s cultivated land yet uses 16% of the world’s insecticides, more than any other single major crop [1]. Pakistan is the world’s fourth largest cotton producer and a major pesticide consumer. Numerous state run organizations have been monitoring the cotton crop for decades through pest-scouting, agriculture surveys and meteorological data-gatherings. This non-digitized, dirty and non-standardized data is of little use for strategic analysis and decision support. An advanced intelligent Agriculture Decision Support System (ADSS) is employed in an attempt to harness the semantic power of that data, by closely connecting visualization and data mining to each other in order to better realize the cognitive aspects of data mining. In this paper, we discuss the critical issue of handling data anomalies of pest scouting data for the six year period: 2001-2006. Using the ADSS it was found that the pesticides were not sprayed based on the pests crossing the critical population threshold, but were instead based on centuries old traditional agricultural significance of the weekday (Monday), thus resulting in non optimized pesticide usage, that can potentially reduce yield.

Heuristics and Meta-heuristics for Bandwidth Minimization of Sparse Matrices

In this paper a new crossing minimization based method is proposed to solve the well-known matrix bandwidth minimization problem, which is to permute the rows and columns of the matrix so as to bring all the non-zero elements of the matrix to reside in a band that is as close as possible to the main diagonal. The feasibility of the crossing minimization paradigm is explored through an objective comparison of four crossing minimization heuristics, along with four meta crossing minimization heuristics, a relatively new clustering technique and two classical bandwidth minimization heuristics i.e. a total of 11 heuristics and meta heuristics. Based on the results of using the simulated data the crossing minimization heuristics are ranked, and the right combination of meta heuristics identified with a surprising finding that meta heuristics take less overall time as compared to the time taken by the corresponding individual heuristics and yet give better results. Lastly all these 11 heuristics and meta heuristics are applied to real data and the outcomes objectively compared with promising results