Optimization Algorithm, Based On The Pruning Of The Ifn Model

With the development of technology in database, rich data have been collected .Knowledge is acquired by means of classification.A lot of methods and technology of classification is used in induction of classification models. The compact and efficient models are given only secondary consideration by most methods and technology of classification Consequently,classification models induced from real-world data tend to be overcomplex and statistically insignificant.In this paper ,the theoretical point which we begin our study is based on the information theory , statistical hypothesis testing and Information-fuzzy Network (IFN) methodology initially introduced by Mark Last and this research aims at building compact and efficient models to analyze datasets. Aim at ignorance of statistical significance IFN statistical hypothesis testing by means of the log likelihood ratio in IFN model, the CIFN(Counting Information-fuzzy Network) algorithm is proposed. The algorithm calculates the mutual information of candidate input attribute and the target attribute given a node.The statistical significance of the mutual information of candidate input attribute and the target attribute is evaluated by using the log likelihood-ratio statistic. The CIFN introduces the threshold of the number of records CIFN each node of given layer so as to guarantee reliability of testing , The CIFN algorithm has the characteristic to construct a compact and efficient data analysis. model .the CIFN algorithm is capable of reducing data dimensionality and is statistically significant.This study makes an efficient exploration in Information-fuzzy Network (IFN) methodology and provides a favorable groundwork to make further researches on data analysis.And the CIFN algorithm has the ability to build an efficient predictive model to help decision maker to manage.