Abu Sayed Md. Latiful Hoque

Clustering medical data to predict the likelihood of diseases

Several studies show that background knowledge of a domain can improve the results of clustering algorithms. In this paper, we illustrate how to use the background knowledge of medical domain in clustering process to predict the likelihood of diseases. To find the likelihood of diseases, clustering has to be done based on anticipated likelihood attributes with core attributes of disease in data point. To find the likelihood of diseases, we have proposed constraint k-Means-Mode clustering algorithm. Attributes of Medical data are both continuous and categorical. The developed algorithm can handle both continuous and discrete data and perform clustering based on anticipated likelihood attributes with core attributes of disease in data point. We have demonstrated its effectiveness by testing it for a real world patient data set.

Storage and Querying of High Dimensional Sparsely Populated Data in Compressed Representation

Storage and querying of high dimensional sparsely populated data creates new challenge to conventional horizontal model. It requires supporting large number of columns and frequently altering of database schema. The sparsity of data degrades performance in both time and space. A 3-ary vertical representation [5] can be used. But the cardinality of the vertical table grows exponentially when the density of the non-null values increases. It is also difficult to support multiple data types usinga single vertical table. In this paper, we have presented a compressed 1-ary vertical representation where schema evolution is easy and size grows linearly with non-null density. Queries can be processed on compressed form of data without decompression. Decompression is done only when the result is necessary. We have considered three alternative representations: 3-ary uncompressed vertical, 1-ary compressed bit-array and 1-ary compressed offset. Experimental results show the superiority of 1-ary offset representation in both space and time.

Towards Development of National Health Data Warehouse for Knowledge Discovery

Availability of accurate data on time is essential for medical decision making. Healthcare organizations own a large amount of data in various systems. Researchers, health care providers and patients will not be able to utilize the knowledge in different stores unless integration of the information from disparate sources is completed. Developing health data warehouse is a complex process and also consumes a significant amount of time but it is essential to deliver quality health services. In this paper the architecture of a data warehouse model and the development process suitable for integrating data from different healthcare sources have been presented. We have developed a Star schema suitable for large data warehouse. Integrating health data requires a rigorous preprocessing and we have completed the preprocessing of national health data by applying efficient transformation techniques. Finally the knowledge discovery potentials from the data warehouse are also presented with relevant examples.

Database Compression Using an Offline Dictionary Method

Off-line dictionary compression is becoming more attractive for applications where compressed data are searched directly in compressed form. While there has been large body of related work describing specific database compression algorithms, the Hibase [10] architecture is unique in processing queries in compressed data. However, this technique does not compress the representation of strings in the domain dictionaries. Primary keys, data with high cardinality and semi-structured data contribute very little or no compression. To achieve high performance irrespective of type of data, the string representation must be in compressed form. At the same time, the direct addressability of compressed data is maintained. Serial compression techniques cannot be used. In this paper, we present a prefix dictionary-based off-line method that can be incorporated with systems like Hibase where compressed data can be accessed directly without prior decompression. The complexity is O(n) in time and space.

Towards development of health Data Warehouse: Bangladesh perspective

Health informatics is one of the top most focuses of researchers now a days. Availability of timely and accurate data is essential for informed medical decision making. Health care organizations face a common problem with large amount of data they have in numerous systems. Such systems are unstructured and unorganized, requires computational time for data integration. Researchers, medical practitioners, health care providers and patients will not be able to utilize the knowledge stored in different repositories unless synthesize the information from disparate sources. This problem can be solved by Data warehousing. Data warehousing techniques share a common set of tasks, include requirements analysis, data design, architectural design, implementation and deployment. Developing Clinical data warehouse is complex and time consuming but is essential to deliver quality patient care. Data integration tasks of medical data store are much challenging when designing clinical data warehouse architecture. This research identifies prospects and complexities of Health data warehousing and Mining in Bangladesh perspective and proposes a data-warehousing model suitable for integrating data from different health care sources.

Equivalence of Problems in Problem Based e-Learning of Database

Problem-based learning (PBL) is a way to learn what is needed to solve a problem, how can a solution be obtained quickly, precisely and professionally. To achieve the goal of problem-based learning, problem design and assign same level of problems among the students are important in engineering classroom environment. SQL is a major part in Database course. In problem based e-Learning of SQL, it is essential to find out the equivalence of an SQL problems to assign the set of problems to a set of students. This is necessary for equal judgment of the performance of individual students. We have developed a complexity model to find out the equivalence of problems for Problem based e-learning of database. In this model, complexity of problems is found by parsing the given solution of the problem in top down approach. We have applied our model to well-known SQL Learning and Evaluation System (SQL-LES). We have compared our calculated complexity value with the complexity value in the question bank of SQL-LES assigned by the SQL experts and found that in most cases our model generate similar complexity value as SQL-LES. Application of our model will reduce the instructor workload in SQL-LES.

Optimized column-oriented model: a storage and search efficient representation of medical data

Medical data have a number of unique characteristics like data sparseness, high dimensionality and rapidly changing set of attributes. Entity Attribute Value (EAV) is the widely used solution to handle the above challenges of medical data, but EAV is neither storage efficient nor search efficient. In this paper, we have proposed a storage & search efficient data model: Optimized Column-Oriented Model (OCOM) for physical representation of high dimensional and sparse data as an alternative of widely used EAV. We have implemented both EAV and OCOM models in a medical data warehousing environment and performed different relational and warehouse queries on both the models. The experimental results show that OCOM is dramatically search efficient and occupy less storage space compared to EAV.

Mining irregular association rules based on action & non-action type data

Conventional positive association rules are the patterns that occur frequently together. These patterns represent what decisions are routinely made based on a set of facts. Irregular association rules are the patterns that represent what decisions are rarely made based on the same set of facts. Many domains like Healthcare, Banking etc need the irregular rule to improve their system. In this paper, we propose a level wise search algorithm that works based on action and non-action type data to find irregular association rules. We have observed that irregular association rules can be discovered efficiently based on action type and non-action type data from large database. To the best of our knowledge, there is no algorithm that can determine such type of associations. Its effectiveness has been demonstrated by testing it for a real world patient data set.

High Performance SQL Queries on Compressed Relational Database

Loss-less data compression is potentially attractive in database application for storage cost reduction and performance improvement. The existing compression architectures work well for small to large database and provide good performance. But these systems can execute a limited number of queries executed on single table. We have developed a disk-based compression architecture, called DHIBASE, to support large database and at the same time, perform high performance SQL queries on single or multiple tables in compressed form. We have compared our system with widely used Microsoft SQL Server. Our system performs significantly better than SQL Server in terms of storage requirement and query response time. DHIBASE requires 10 to 15 times less space and for some operation it is 18 to 22 times faster. As the system is column oriented, schema evolution is easy.

A storage & search efficient representation of medical data

Integrating Large-scale medical data is important to support healthcare research. Medical data have a number of unique characteristics like data sparseness, high dimensionality and rapidly changing set of attributes. To handle the above challenges of medical data, Entity Attribute Value (EAV) is the widely used solution. However, EAV suffers from higher storage requirement and not search efficient. In this paper, we have proposed a storage & search efficient data model: Optimized Column-Oriented Model (OCOM) for the physical representation of Medical data as an alternative of widely used EAV. This model optimizes basic Column-Oriented model. We have demonstrated its effectiveness by testing it in a medical data-warehousing environment.