Search This Blog

Thursday, April 14, 2011

KDD-Knowledge Discovery in Database

Knowledge Discovery in Databases
There are almost as many differing definitions of the term ''Data Mining'' as there are authors who have written about it. Since Data Mining sits at the interface of a variety of fields, e.g. computer science, statistics, artificial intelligence, business information systems, and machine learning, its definition changes with the field's perspective. Computer scientists, typically, refer to Data Mining as a clearly defined part of the Knowledge Discovery in Databases (KDD) process, while many statisticians use Data Mining as a synonym for the whole KDD process.
To get a flavor of both the variation as well as the common core of data and knowledge mining, we cite some of the definitions used in the literature.
KDD is the non-trivial process of identifying valid, novel, potentially useful, and ultimately understandable patterns in data.

([8])


Knowledge discovery is a knowledge-intensive task consisting of complex interactions, protracted over time, between a human and a (large) database, possibly supported by a heterogenous suite of tools.

([4])


[Data Mining is] a step in the KDD process consisting of particular data mining algorithms that, under some acceptable computational efficiency limitations, produce a particular enumeration of patterns.

([8])


[Data Mining is] a folklore term which indicates application, under human control, of low-level data mining methods. Large scale automated search and interpretation of discovered regularities belong to KDD, but are typically not considered part of data mining.

([24])


[Data Mining is] used to discover patterns and relationships in data, with an emphasis on large observational data bases. It sits at the common frontiers of several fields including Data Base Management, Artificial Intelligence, Machine Learning, Pattern Recognition, and Data Visualization.

([10])


[Data Mining is] the process of secondary analysis of large databases aimed at finding unsuspected relationships which are of interest or value to the database owners.

([12])
Data Mining is the analysis of (often large) observational data sets to find unsuspected relationships and to summarize the data in novel ways that are both understandable and useful to the data owner.

([13])
From these definitions the essence is that we are talking about exploratory analysis of large data sets. Two further aspects are the use of computer-based methods and the notion of secondary and observational data. The latter means that the data do not come from experimental studies and that data was originally collected for some other purpose, either for a study with different goals or for record-keeping reasons. These four characteristics in combination distinguish the field of Data Mining from traditional statistics. The exploratory approach in Data Mining clearly defines the goal of finding patterns and generating hypothesis, which might later on be subject of designed experiments and statistical tests. Data sets can be large at least in two different aspects. The most common one is in form of a large number of observations (cases). Real world applications usually are also large in respect of the number of variables (dimensions) that are represented in the data set. Data Mining is also concerned with this side of largeness. Especially in the field of bioinformatics, many data sets comprise only a small number of cases but a large number of variables. Secondary analysis implies that the data can rarely be regarded as a random sample from the population of interest and may have quite large selection biases. The primary focus in investigating large data sets tends not to be the standard statistical approach of inferencing from a small sample to a large universe, but more likely partitioning the large sample into homogeneous subsets.
The ultimate goal of Data Mining methods is not to find patterns and relationships as such, but the focus is on extracting knowledge, on making the patterns understandable and usable for decision purposes. Thus, Data Mining is the component in the KDD process that is mainly concerned with extracting patterns, while Knowledge Mining involves evaluating and interpreting these patterns. This requires at least that patterns found with Data Mining techniques can be described in a way that is meaningful to the data base owner. In many instances, this description is not enough, instead a sophisticated model of the data has to be constructed.
Data pre-processing and data cleansing is an essential part in the Data and Knowledge Mining process. Since data mining means taking data from different sources, collected at different time points, and at different places, integration of such data as input for data mining algorithms is an easily recognized task, but not easily done. Moreover, there will be missing values, changing scales of measurement, as well as outlying and erroneous observations. To assess the data quality is a first and important step in any scientific investigation. Simple tables and statistical graphics give a quick and concise overview on the data, to spot data errors and inconsistencies as well as to confirm already known features. Besides the detection of uni- or bivariate outliers graphics and simple statistics help in assessing the quality of the data in general and to summarize the general behavior. It is worth noting that many organizations still report that as much as $ 80\,{\%}$ of their effort for Data and Knowledge Mining goes into supporting the data cleansing and transformation process.

No comments:

Post a Comment