Feature Extraction and Reduction by using Modified Apriori algorithm

Feature selection is of great importance in the world of data mining. The existing algorithms for doing so lead to extremely high dimensionality when interactions among features are considered. So, an algorithm is needed to extract only useful features. An Improvised approach for association rule mining is devised, which does not require the users to provide support and confidence values and we are considering the interaction among features and relative confidence to get the better result. The user can specify the number of rules and features that have to be generated. This method has an advantage over Standard Apriori, as it gives more control over the efficiency and quality of the results. It is also capable to handle large data sets and high dimensions more effectively. We calculated mean, median, variance and standard deviation on the lift value of rules generated and found that variance of improvised apriori is 30 – 40% lower than that of standard apriori, which indicates that modified Apriori performs better than standard apriori. The mean and median value of the lift is also 20 – 50% higher in modified apriori than in standard apriori, which implies that the features are more tightly coupled in modified apriori than in standard apriori and hence have a higher probability of getting the perfect result