Feature selection method and device

Abstract

The invention provides a feature selection method and device. The method includes the first step of generating a first training sample set, a first feature index set corresponding to the first training sample set, a second training sample set and a second feature index set corresponding to the second training sample set through response to received training sample sets, the second step of calculating a first element according to the first training sample set and completing updating of the first feature index set, the third step of calculating a second element according to the second training sample set and completing updating of the second feature index set, and the fourth step of calculating a feature index set according to the obtained first feature index set and the obtained second feature index set when the sum of the number of all features in the updated first feature index set and the number of all features in the updated second feature index set reaches a preset value, and then completing selection of the features. Consequently, withholding calculation is reduced, and the generalization capability is improved in the feature selection process on the basis of ensuring the learning efficiency.

Description

technical field [0001] The present application relates to the technical field of medical diagnosis, and in particular, to a feature selection method and device. Background technique [0002] In today's society, any field is inseparable from the help of computers. The same is true in the field of medical diagnosis. It uses some technologies to simulate the diagnosis and treatment of diseases by medical experts, which can effectively solve various clinical problems and play the role of "doctor's assistant", especially to help young and inexperienced doctors improve their diagnostic skills. , optimize the diagnosis and treatment plan. The application of machine learning in medical diagnosis is from the most primitive application of prior knowledge for heuristic reasoning, to the later probability calculation method and artificial intelligence method, and then to the widely used neural network technology, simulation technology and genetic algorithm, etc. Learning is increasing...

AI Technical Summary

Problems solved by technology

[0004] 1. Filter (filtering) method, the feature selection process of this method has nothing to do with the learning algorithm, it is to estimate the effectiveness of a certain feature subset through the value of an adaptive function, and has nothing to do with the specific classifier, although this method can be used independently learning algorithm, but its learning efficiency is not high

[0005] 2. Wrapper (encapsulation) method. The feature selection process of this method is related to the learning algorithm. It uses the performance of a specific classifier as the criterion for feature subset selection. Although this strategy of directly optimizing the classifier can improve the classification The generalization of the machine can improve the learning efficiency, but its calculation and payment are relatively high, and the promotion ability is poor

Images