Feature selection method

Abstract

The invention provides a feature selection method which comprises the following steps of: (a) inputting an entire feature set serving as a complete feature space into a feature selection algorithm; (b) taking features input into the feature space according to a certain sequence and putting the features into a new feature space; (c) comparing the features in different spaces by using a feature evaluation criterion; (d) repeating the steps (b) and (c) until the number of the features in the new space reaches a default value; and (e) considering the features in the new feature space as the selected optimal feature subset. By adopting the method, setting of a threshold value is saved, the number of the features in the result can be selected, the time of an entire selection process can be conveniently controlled, and a better feature subset can be selected.

Description

technical field [0001] The invention relates to a feature selection method. Background technique [0002] Feature selection is one of the three cores of pattern recognition. It is widely used in the fields of artificial intelligence, pattern recognition, image processing and target recognition. As the purpose of various applications changes from single to multiple, the use environment changes from simple to complex, different usage requirements and the number of features used by applications continue to increase, feature selection is becoming more and more important and should be more flexible. [0003] Ron Kohavi and George H.John.Wrappers For Feature Subset Selection (Artificial Intelligence 97.1997, 273-275, 283-286) proposed a feature extraction model, which is called an encapsulated feature extraction method. Compared with other feature selection methods, encapsulated feature selection adds the part of using subsequent learning algorithm to identify feature subsets. ...

AI Technical Summary

Problems solved by technology

But on the one hand, it is difficult to determine ε in practical applications, whether the value of ε is too large or too small will have a great impact on the performance of the algorithm; on the other hand, due to the uncertain size of the feature space, the number of selected features will Not sure, but in practical applications, the time-consuming of packaged feature selection increases with the number of selected features. If the selected best feature space is too large, the time overhead of the selection process will be very large, which is inconvenient. practical application

Images