Pattern Discovery:An Inference Analysis Approach

Data mining is a process of discovering knowledge of interest for the users from various information repositories like databases and data warehouses by extracting interesting data patterns that represent knowledge. These interesting data patterns are obtained by evaluating the existing data patterns based on some interestingness measures. This data mining process is carried out by using several data mining functionalities among which Association Rule Mining is most commonly used to find interesting data patterns based on the association relationship among various data items of a data set. This association relation is represented by an association rule, which is a data pattern. To evaluate the interestingness of an association rule, two interestingness measures called as support and confidence are used whose threshold values are set by domain experts. These threshold values called as minimum support and minimum confidence which represent domain knowledge. These threshold values cannot be always accurate, which leads to loss of interesting association rules and also affects the quality of association rules discovered. This leads a user to take wrong decisions.
To avoid this problem, we need to move from this support-confidence framework for a process which can find interesting association rules based on their logical correctness. An association rule is logically correct only when it satisfies a logical principle called Equivalence, such rules are called as coherent rules. As these coherent rules are discovered without using domain knowledge, no interesting association rules are lost. Though this process is accurate, it is a time consuming task to check the logical correctness of every association rule. So we need to use a technique called Pruning, through which we can get accurate results by consuming less time.